Download Pinker - Bloom - NATURAL LANGUAGE AND NATURAL SELECTION

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Steven Pinker wikipedia , lookup

Transcript
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 1
NATURAL LANGUAGE AND NATURAL SELECTION
Steven Pinker and Paul Bloom
Keywords
Language, Evolution, Language Acquisition, Natural Selection, Grammatical
Theory, Biology of Language, Language Universals, Psycholinguistics, Origin of
Language
Abstract
Many people have argued that the evolution of the human language faculty cannot
be explained by Darwinian natural selection. Chomsky and Gould have suggested
that language may have evolved as the by-product of selection for other abilities or
as a consequence of as-yet unknown laws of growth and form. Others have argued
that a biological specialization for grammar is incompatible with every tenet of
Darwinian theory -- that it shows no genetic variation, could not exist in any
intermediate forms, confers no selective advantage, and would require more
evolutionary time and genomic space than is available. We examine these
arguments and show that they depend on inaccurate assumptions about biology or
language or both. Evolutionary theory offers clear criteria for when a trait should be
attributed to natural selection: complex design for some function, and the absence
of alternative processes capable of explaining such complexity. Human language
meets this criterion: grammar is a complex mechanism tailored to the transmission
of propositional structures through a serial interface. Autonomous and arbitrary
grammatical phenomena have been offered as counterexamples to the position
that language is an adaptation, but this reasoning is unsound: communication
protocols depend on arbitrary conventions that are adaptive as long as they are
shared. Consequently, language acquisition in the child should systematically differ
from language evolution in the species and attempts to analogize them are
misleading. Reviewing other arguments and data, we conclude that there is every
reason to believe that a specialization for grammar evolved by a conventional neoDarwinian process.
Language could not have begun in the form it was said to have taken
in the first recorded utterance of Thomas Babbington Macaulay (the
infant Lord Macaulay): once when he was taken out, his hostess
accidently spilled hot tea on him. The little lad first bawled his head
off, but when he had calmed he said in answer to his hostess'
concern, "Thank you Madam, the agony is sensibly abated." -- P. B.
and J. S. Medawar
1. Introduction
All human societies have language. As far as we know they always did; language
was not invented by some groups and spread to others like agriculture or the
alphabet. All languages are complex computational systems employing the same
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 2
basic kinds of rules and representations, with no notable correlation with
technological progress: the grammars of industrial societies are no more complex
than the grammars of hunter-gatherers; Modern English is not an advance over
Old English. Within societies, individual humans are proficient language users
regardless of intelligence, social status, or level of education. Children are fluent
speakers of complex grammatical sentences by the age of three, without benefit of
formal instruction. They are capable of inventing languages that are more
systematic than those they hear, showing resemblances to languages that they
have never heard, and they obey subtle grammatical principles for which there is
no evidence in their environments. Disease or injury can make people linguistic
savants while severely retarded, or linguistically impaired with normal intelligence.
Some language disorders are genetically transmitted. Aspects of language skill can
be linked to characteristic regions of the human brain. The human vocal tract is
tailored to the demands of speech, compromising other functions such as
breathing and swallowing. Human auditory perception shows complementary
specializations toward the demands of decoding speech sounds into linguistic
segments.
This list of facts (see Pinker, 1989a) suggests that the ability to use a natural language
belongs more to the study of human biology than human culture; it is a topic like
echolocation in bats or stereopsis in monkeys, not like writing or the wheel. All modern
students of language agree that at least some aspects of language are due to speciesspecific, task-specific biological abilities, though of course there are radical disagreements
about specifics. A prominent position, outlined by Chomsky (1965, 1980, 1981, 1986,
1988a), Fodor (1983), Lenneberg (1964, 1967), and Liberman (Liberman, Cooper,
Shankweiler, & Studdert-Kennedy, 1967; Liberman and Mattingly, 1989), is that the mind
is composed of autonomous computational modules -- mental faculties or "organs" -- and
that the acquisition and representation of language is the product of several such specialized
modules.
It would be natural, then, to expect everyone to agree that human language is the product of
Darwinian natural selection. The only successful account of the origin of complex
biological structure is the theory of natural selection, the view that the differential
reproductive success associated with heritable variation is the primary organizing force in
the evolution of organisms (Darwin, 1859; see Bendall, 1983 for a contemporary
perspective). But surprisingly, this conclusion is contentious. Noam Chomsky, the world's
best-known linguist, and Stephen Jay Gould, the world's best-known evolutionary theorist,
have repeatedly suggested that language may not be the product of natural selection, but a
side effect of other evolutionary forces such as an increase in overall brain size and
constraints of as-yet unknown laws of structure and growth (e.g., Chomsky, 1972, 1982a,
1982b, 1988a, 1988b; Gould, 1987a; Gould and Piattelli-Palmarini, 1987). Recently
Massimo Piattelli-Palmarini (1989), a close correspondent with Gould and Chomsky, has
done the field a service by formulating a particularly strong version of their positions and
articulating it in print. Premack (1985, 1986) and Mehler (1985) have expressed similar
views.
In this paper we will examine this position in detail, and will come to a very different
conclusion. We will argue that there is every reason to believe that language has been
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 3
shaped by natural selection as it is understood within the orthodox "synthetic" or "neoDarwinian" theory of evolution (Mayr, 1982). In one sense our goal is incredibly boring.
All we argue is that language is no different from other complex abilities such as
echolocation or stereopsis, and that the only way to explain the origin of such abilities is
through the theory of natural selection. One might expect our conclusion to be accepted
without much comment by all but the most environmentalist of language scientists (as
indeed it is by such researchers as Bickerton, 1981, Liberman and Mattingly, 1989,
Lieberman, 1984, and, in limited respects, by Chomsky himself in some strands of his
writings.(Note 1)). On the other hand, when two such important scholars as Chomsky and
Gould repeatedly urge us to consider a startling contrary position, their arguments can
hardly be ignored. Indeed these arguments have had a strong effect on many cognitive
scientists, and the nonselectionist view has become the consensus in many circles.
Furthermore, a lot is at stake if our boring conclusion is wrong. We suspect that many
biologists would be surprised at the frequent suggestion that the complexity of language
cannot be explained through natural selection. For instance, Chomsky has made the
following statements:
[an innate language faculty] poses a problem for the biologist, since,
if true, it is an example of true 'emergence' -- the appearance of a
qualitatively different phenomenon at a specific stage of complexity of
organization. (1972: 70)
It is perfectly safe to attribute this development [of innate mental structure]
to "natural selection", so long as we realize that there is no substance to this
assertion, that it amounts to nothing more than a belief that there is some
naturalistic explanation for these phenomena. (1972: 97)
Evolutionary theory is informative about many things, but it has little to say,
as of now, of questions of this nature [e.g., the evolution of language]. The
answers may well lie not so much in the theory of natural selection as in
molecular biology, in the study of what kinds of physical systems can
develop under the conditions of life on earth and why, ultimately because of
physical principles. (1988a: 167)
It does seem very hard to believe that the specific character of organisms can
be accounted for purely in terms of random mutation and selectional
controls. I would imagine that the biology of a 100 years from now is going
to deal with the evolution of organisms the way it now deals with the
evolution of amino acids, assuming that there is just a fairly small space of
physically possible systems that can realize complicated structures. ..
Evolutionary theory appears to have very little to say about speciation, or
about any kind of innovation. It can explain how you get a different
distribution of qualities that are already present, but it does not say much
about how new qualities can emerge. (1982a:23)
If findings coming out of the study of language forced biologists to such
conclusions, it would be big news.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 4
There is another reason to scrutinize the nonselectionist theory of language. If a current
theory of language is truly incompatible with the neo-Darwinian theory of evolution, one
could hardly blame someone for concluding that it is not the theory of evolution that must
be questioned, but the theory of language. Indeed, this argument has been the basis of
critiques of Chomsky's theories by Bates, Thal, and Marchman (1989), Greenfield (1987),
and Lieberman (1984, 1989), who are nonetheless strange bedfellows with Chomsky in
doubting whether an innate generative grammar could have evolved by natural selection.
Since we are impressed both by the synthetic theory of evolution and by the theory of
generative grammar, we hope that we will not have to choose between the two.
In this paper, we first examine arguments from evolutionary biology about when it is
appropriate to invoke natural selection as an explanation for the evolution of some trait. We
then apply these tests to the case of human language, and conclude that language passes.
We examine the motivations for the competing nonselectionist position, and suggest that
they have little to recommend them. In the final section, we refute the arguments that have
claimed that an innate specialization for grammar is incompatible with the tenets of a
Darwinian account and thus that the two are incompatible.
2. The Role of Natural Selection in Evolutionary Theory
Gould has frequently suggested that evolutionary theory is in the throes of a
scientific revolution (e.g., Eldredge & Gould, 1977; Gould, 1980). Two cornerstones
of the Darwinian synthesis, adaptationism and gradualism, are, he argues, under
challenge. Obviously if strict Darwinism is false in general it should not be used to
explain the origin of language.
2.1. Nonselectionist Mechanisms of Evolutionary Change
In a classic paper, Gould and Lewontin (1979) warn against "naive adaptationism,"
the inappropriate use of adaptive theorizing to explain traits that have emerged for
other reasons (see also Kitcher, 1983; Lewontin, 1978). The argument is illustrated
by an analogy with the mosaics on the dome and spandrels of the San Marco
basilica in Venice:
Spandrels -- the tapering triangular spaces formed by the intersection
of two rounded arches at right angles ... are necessary architectural
by-products of mounting a dome on rounded arches. Each spandrel
contains a design admirably fitted into its tapering space. An
evangelist sits in the upper part flanked by the heavenly cities. Below,
a man representing one of the four biblical rivers ... pours water from
a pitcher in the narrowing space below his feet.
The design is so elaborate, harmonious, and purposeful that we are tempted
to view it as the starting point of any analysis, as the cause in some sense of
the surrounding architecture. But this would invert the proper path of
analysis. The system begins with an architectural constraint: the necessary
four spandrels and their tapering triangular form. They provide a space in
which the mosaicists worked; they set the quadripartite symmetry of the
dome above.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 5
Such architectural constraints abound, and we find them easy to understand
because we do not impose our biological biases upon them. ... Anyone who
tried to argue that the structure [spandrels] exists because of [the designs laid
upon them] would be inviting the same ridicule that Voltaire heaped on Dr.
Pangloss: "Things cannot be other than they are ... Everything is made for
the best purpose. Our noses were made to carry spectacles, so we have
spectacles. Legs were clearly intended for breeches, and we wear them." ...
Yet evolutionary biologists, in their tendency to focus exclusively on
immediate adaptation to local conditions, do tend to ignore architectural
constraints and perform just such an inversion of explanation. (pp. 147-149)
Unconvincing adaptationist explanations, which Gould and Lewontin compare to
Kipling's "Just-so stories," are easy to find. In the Science and Technology section
of the Boston Globe in March 1987, an article noted that the number of teats in
different mammals ought to correspond not to the average litter size but to the
largest litter size that can occur for that species within some bound of probability.
Since humans ordinarily bear single children but not infrequently have twins, we
have an explanation for why humans have two breasts, not one. The author did not
discuss the possibility that the bilateral symmetry that is so basic to the mammalian
body plan makes the appearance of one-breasted humans rather unlikely. Gould
and Lewontin describe a number of nonadaptationist mechanisms that they feel
are frequently not tested within evolutionary accounts: genetic drift, laws of growth
and form (such as general allometric relations between brain and body size), direct
induction of form by environmental forces such as water currents or gravity, the
effects of accidents of history (which may trap organisms in local maxima in the
adaptive landscape), and "exaptation" (Gould and Vrba, 1982), whereby new uses
are made of parts that were originally adapted to some other function or of
spandrels that had no function at all but were present for reasons of architecture,
development, or history. They point out that Darwin himself had this pluralistic view
of evolution, and that there was an "unfairly maligned" nonadaptationist approach
to evolution, prominent in continental Europe, that stressed constraints on
"Baupl@act[c]ne" (architectural plans) flowing from phyletic history and
embryological development. This body of research, they suggest, is an antidote to
the tendency to treat an organism as a bundle of traits or parts, each independently
shaped by natural selection.
2.2. Limitations on Nonselectionist Explanations
The Gould and Lewontin argument could be interpreted as stressing that since the
neo-Darwinian theory of evolution includes nonadaptationist processes it is bad
scientific practice not to test them as alternatives to natural selection in any
particular instance. However, they are often read as having outlined a radical new
alternative to Darwin, in which natural selection is relegated to a minor role.
Though Gould and Lewontin clearly eschew this view in their paper, Gould has
made such suggestions subsequently (e.g., Gould, 1980), and Piattelli-Palmarini
(1989: 1) has interpreted it as such when he talks of Darwinian natural selection
being replaced by "a better evolutionary theory (one based on 'exaptation')". The
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 6
reasons why we should reject this view were spelled out clearly by Williams (1966),
and have been amplified recently by Dawkins (1983, 1986).
The key point that blunts the Gould and Lewontin critique of adaptationism is that natural
selection is the only scientific explanation of adaptive complexity. "Adaptive complexity"
describes any system composed of many interacting parts where the details of the parts'
structure and arrangement suggest design to fulfill some function. The vertebrate eye is the
classic example. The eye has a transparent refracting outer cover, a variable-focus lens, a
light-sensitive layer of neural tissue lying at the focal plane of the lens, a diaphragm whose
diameter changes with illumination level, muscles that move it in precise conjunction and
convergence with those of the other eye, and elaborate neural circuits that respond to
patterns defining edges, colors, motion, and stereoscopic disparity. It is impossible to make
sense of the structure of the eye without noting that it appears as if it was designed for the
purpose of seeing -- if for no other reason that the man-made tool for image formation, the
camera, displays an uncanny resemblance to the eye. Before Darwin, theologians, notably
William Paley, pointed to its exquisite design as evidence for the existence of a divine
designer. Darwin showed how such "organs of extreme perfection and complication" could
arise from the purely physical process of natural selection.
The essential point is that no physical process other than natural selection can explain the
evolution of an organ like the eye. The reason for this is that structures that can do what the
eye does are extremely low-probability arrangements of matter. By an unimaginably large
margin, most objects defined by the space of biologically possible arrangements of matter
cannot bring an image into focus, modulate the amount of incoming light, respond to the
presence of edges and depth boundaries, and so on. The odds that genetic drift, say, would
result in the fixation within a population of just those genes that would give rise to such an
object are infinitesimally small, and such an event would be virtually a miracle. This is also
true of the other nonselectionist mechanisms outlined by Gould and Lewontin. It is
absurdly improbable that some general law of growth and form could give rise to a
functioning vertebrate eye as a by-product of some other trend such as an increase in size of
some other part. Likewise, one need not consider the possibility that some organ that arose
as an adaptation to some other task, or a spandrel defined by other body parts, just
happened to have a transparent lens surrounded by a movable diaphragm in front of a lightsensitive layer of tissue lying at its focal plane. Natural selection -- the retention across
generations of whatever small, random modifications yield improvements in vision that
increase chances of survival and reproduction -- is the only physical process capable of
creating a functioning eye, because it is the only physical process in which the criterion of
being good at seeing can play a causal role. As such it is the only process that can lead
organisms along the path in the astronomically vast space of possible bodies leading from a
body with no eye to a body with a functioning eye.
This argument is obviously incomplete, as it relies on the somewhat intuitive notion of
"function" and "design." A skeptic might accuse the proponent of circularity, asking why a
lump of clay should not be considered well-designed to fulfill the function of taking up
exactly the region of space that it in fact takes up. But the circle can be broken in at least
three ways. First, biologists need posit far fewer functions than there are biological
systems; new functions are not invented for each organ of each organism. Furthermore,
each legitimate function can be related via a direct plausible causal chain to other functions
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 7
and -- critically -- to the overall function of survival and reproduction. Finally, convergent
evolution and resemblance to human artifacts fulfilling the same putative function give
independent criteria for design. But regardless of the precise formulation of the modern
argument from design (see, e.g., Cummins, 1984), it is not controversial in practice. Gould
himself readily admits that natural selection is the cause of structures such as the vertebrate
eye, and he invokes the criterion of engineering design, for example, to rescue Darwinism
itself from the charge of circularity (Gould, 1977a). Presumably this is why Gould and
Lewontin concede that they agree with Darwin that natural selection is "the most important
of evolutionary mechanisms."
What, then, is the proper relation between selectionist and nonselectionist explanations in
evolution? The least interesting case involves spandrels that are not involved in any
function or behavior, such as the redness of blood, the V-shaped space between a pair of
fingers, the hollow at the back of a knee, the fact there are a prime number of digits on each
limb, and so on. The mere presence of these epiphenomenal spandrels, that play no direct
role in the explanation of any species-typical behavior or function, says nothing about
whether the structures that they are associated with were shaped by selection. There are as
many of them as there are ways of describing an organism that do not correspond to its
functional parts.
Much more important are cases where spandrels are modified and put to use. However, in
such cases of modified spandrels, selection plays a crucial role. Putting a dome on top of
four arches gives you a spandrel, but it does not give you a mosaic depicting an evangelist
and a man pouring water out of a pitcher. That would really be a miracle. To get the actual
mosaic you need a designer. The designer corresponds to natural selection. Spandrels,
exaptations, laws of growth, and so on can explain the basic plans, parts, and materials that
natural selection works with -- as Jacob (1977) put it, nature is a tinkerer, not an engineer
with a clean drawing board. The best examples of structures produced entirely by
nonadaptationist mechanisms are generally one-part or repetitive shapes or processes that
correspond to simple physical or geometric laws, such as chins, hexagonal honeycombs,
large heads on large bodies, and spiral markings. But, as Darwin stressed, when such parts
and patterns are modified and combined into complex biological machines fulfilling some
delicate function, these subsequent modifications and arrangements must be explained by
natural selection.
The real case of evolution without selection consists of the use of unmodified spandrels.
Gould (1987a) describes a kind of wading bird that uses its wings primarily to block
reflections on the surface of water while looking for fish. The possibility that some useful
structure is an unmodified spandrel is the most interesting implication of the GouldLewontin argument, since Darwinian natural selection would really play no role. Note,
though, that unmodified spandrels have severe limitations. A wing used as a visor is a case
where a structure designed for a complex engineering task that most arrangements of matter
do not fulfill, such as controlled flight, is exapted to a simple engineering task that many
arrangements of matter do fulfill, such as screening out reflections (we are reminded of the
paperweight and aquarium depicted in 101 Uses for a Dead Computer.) When the reverse
happens, such as when a solar heat exchanger is retooled as a fully functioning wing in the
evolution of insects (Kingsolver and Koehl, 1985), natural selection must be the cause.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 8
We are going over these criteria for invoking natural selection in such detail because they
are so often misunderstood. We hope we have made it clear why modern evolutionary
biology does not license Piattelli-Palmarini's conclusion that "since language and cognition
probably represent the most salient and the most novel biological traits of our species, ... it
is now important to show that they may well have arisen from totally extra-adaptive
mechanisms." And Piattelli-Palmarini is not alone. In many discussions with cognitive
scientists we have found that adaptation and natural selection have become dirty words.
Anyone invoking them is accused of being a a naive adaptationist, or even of
"misunderstanding evolution." Worst of all, he or she is open to easy ridicule as a Dr.
Pangloss telling Just-so stories! (Premack's 1986 reply to Bickerton, 1986, is typical.)
Given the uncontroversially central role of natural selection in evolution, this state of affairs
is unfortunate. We suspect that many people have acquired much of their knowledge of
evolutionary theory from Gould's deservedly popular essays. These essays present a view
of evolution that is vastly more sophisticated than the 19th-Century versions of Darwin
commonly taught in high schools and even colleges. But Gould can easily be misread as
fomenting a revolution rather than urging greater balance within current biological
research, and his essays do not emphasize the standard arguments for when it is
appropriate, indeed necessary, to invoke natural selection.
Also lurking beneath people's suspicions of natural selection is a set of methodological
worries. Isn't adaptationism fundamentally untestable, hence unscientific, because adaptive
stories are so easy to come by that when one fails, another can always be substituted?
Gould and Lewontin may be correct in saying that biologists and psychologists have leapt
too quickly to unmotivated and implausible adaptationist explanations, but this has nothing
to do with the logic of adaptationist explanations per se. Glib, unmotivated proposals can
come from all kinds of theories. To take an example close to home, the study of the
evolution of language attained its poor reputation precisely because of the large number of
silly nonadaptationist hypotheses that were proposed. For instance, it has been argued that
language arose from mimicry of animal calls, imitations of physical sounds, or grunts of
exertion (the infamous "bow-wow," "ding-dong," and "heave-ho" theories.)
Specific adaptationist proposals are testable in principle and in practice (see Dennett, 1983;
Kitcher, 1983; Maynard Smith, 1984, Mayr, 1982; Sober, 1984; Williams, 1966.)
Supplementing the criterion of complex design, one can determine whether putatively
adaptive structures are correlated with the ecological conditions that make them useful, and
under certain circumstance one can actually measure the reproductive success of
individuals possessing them to various degrees (see, e.g., Clutton-Brock, 1983). Of course,
the entire theory of natural selection may be literally unfalsifiable in the uninteresting sense
that elaborations can always rescue its empirical failings, but this is true of all large-scale
scientific theories. Any such theory is supported to the extent that the individual
elaborations are mutually consistent, motivated by independent data, and few in number
compared to the phenomena to be explained.(Note 2)
Indeed one could argue that it is nonadaptationist accounts that are often in grave danger of
vacuity. Specific adaptationist proposals may be unmotivated, but they are within the realm
of biological and physical understanding, and often the problem is simply that we lack the
evidence to determine which account within a set of alternative adaptive explanations is the
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 9
correct one. Nonadaptationist accounts that merely suggest the possibility that there is some
hitherto-unknown law of physics or constraint on form -- a "law of eye-formation," to take
a caricatured example -- are, in contrast, empty and nonfalsifiable.
2.3. Two Issues that are Independent of Selectionism
There are two other issues that Gould includes in his depiction of a scientific
revolution in evolutionary theory. It is important to see that they are largely
independent of the role of selection in evolutionary change.
2.3.1. Gradualism
According to the theory of "punctuated equilibrium" (Eldredge and Gould, 1972;
Gould and Eldredge, 1977), most evolutionary change does not occur continuously
within a lineage, but is confined to bursts of change that are relatively brief on the
geological time scale, generally corresponding to speciation events, followed by
long periods of stasis. Gould has suggested that the theory has some very general
and crude parallels with approaches to evolution that were made disreputable by
the neo-Darwinian synthesis, approaches that go by the names of "saltationism,"
"macromutations," or "hopeful monsters." (e.g., Gould, 1981). However, he is
emphatic that punctuated equilibrium is "a theory about ordinary speciation (taking
tens of thousands of years) and its abrupt appearance at low scales of geological
resolution, not about ecological catastrophe and sudden genetic change" (Gould,
1987b: 234). Many other biologists see evolutionary change in an even more
orthodox light. They attribute the sudden appearance of fully-formed new kinds of
organisms in the fossil record to the fact that speciation typically takes place in
small, geographically isolated populations. Thus transitional forms, even if evolving
over very long time-spans, are unlikely to appear in the fossil record until they
reinvade the ancestral territory; it is only the invasion that is sudden (see, e.g.,
Ayala, 1983; Dawkins, 1986; Mayr, 1982; Stebbins and Ayala, 1981). In any case it
is clear that evolutionary change is gradual from generation to generation, in full
agreement with Darwin. Thus Piattelli-Palmarini (1989: 8) expresses a common
misunderstanding when he interprets the theory of punctuated equilibrium as
showing that "many incomplete series in the fossil record are incomplete, not
because the intermediate forms have been lost for us, but because they simply
never existed".
Once again the explanation of adaptive complexity is the key reason why one should reject
nongradual change as playing an important role within evolution. An important Darwinian
insight, reinforced by Fisher (1930), is that the only way for complex design to evolve is
through a sequence of mutations with small effects. Although it may not literally be
impossible for an organ like the eye to emerge across one generation from no eye at all, the
odds of this happening are unimaginably low. A random large leap in the space of possible
organic forms is astronomically unlikely to land an organism in a region with a fully
formed functioning eye. Only a hill-climbing process, with each small step forced in the
direction of forms with better vision, can guide the lineage to such a minuscule region of
the space of possible forms within the lifetime of the universe.
None of this is to deny that embryological processes can result in quite radical singlegeneration morphological changes. "Homeotic" mutations causing slight changes in the
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 10
timing or positioning of epigenetic processes can result in radically new kinds of offspring,
such as fruit flies with legs growing where their antenna should be, and it is possible that
some speciation events may have begun with such large changes in structure. However
there is a clear sense in which such changes are still gradual, since they only involve a gross
modification or duplication of existing structure, not the appearance of a new kind of
structure (see Dawkins, 1983).
2.3.2. Exaptation
Exaptation is another process that is sometimes discussed as if it was incompatible
both with adaptationism and with gradualism. People often wonder whether each of
the "numerous, successive, slight modifications" from an ancestor lacking an organ
to a modern creature enjoying the fully-functioning organ leads to an improvement
in the function, as it should if the necessary evolutionary sequence is to be
complete. Piattelli-Palmarini cites Kingsolver and Koehl's (1985) study of qualitative
shifts during the evolution of wings in insects, which are ineffective for flight below
a certain size, but effective as solar heat exchange panels precisely within that
range. (The homologies among parts of bat wings, seal flippers, horse forelimbs,
and human arms is a far older example). Nevertheless such exaptations are still
gradual and are still driven by selection; there must be an intermediate evolutionary
stage at which the part can subserve both functions (Mayr, 1982), after which the
process of natural selection shapes it specifically for its current function. Indeed the
very concept of exaptation is essentially similar to what Darwin called
"preadaptation," and played an important role in his explanation of "the incipient
stages of useful structures."
Furthermore, it is crucial to understand that exaptation is merely one empirical possibility,
not a universal law of evolution. Gould is often quoted as saying "We avoid the excellent
question, What good is 5 percent of an eye? by arguing that the possessor of such an
incipient structure did not use it for sight."(1977b:107). (Of course no ancestor to humans
literally had 5 percent of a human eye; the expression refers to an eye that has 5 percent of
the complexity of a modern eye). In response, Dawkins (1986: 81) writes: "An ancient
animal with 5 per cent of an eye might indeed have used it for something other than sight,
but it seems to me at least as likely that it used it for 5 per cent vision. ... Vision that is 5
percent as good as yours or mine is very much worth having in comparison with no vision
at all. So is 1 per cent vision better than total blindness. And 6 per cent is better than 5, 7
per cent better than 6, and so on up the gradual, continuous series." Indeed Darwin (1859)
sketched out a hypothetical sequence of intermediate forms in the evolution of the
vertebrate eye, all with counterparts in living organisms, each used for vision.
In sum, the positions of Gould, Lewontin, and Eldredge should not be seen as radical
revisions of the theory of evolution, but as a shift in emphasis within the orthodox neoDarwinian framework. As such they do not invalidate gradual natural selection as the
driving force behind the evolution of language on a priori grounds. Furthermore, there are
clear criteria for when selectionist and nonselectionist accounts should be invoked to
explain some biological structure: complex design to carry out some reproductively
significant function, versus the existence of a specific physical, developmental or random
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 11
process capable of explaining the structure's existence. With these criteria in hand, we can
turn to the specific problem at hand, the evolution of language.
3. Design in Language
Do the cognitive mechanisms underlying language show signs of design for some
function in the same way that the anatomical structures of the eye show signs of
design for the purpose of vision? This breaks down into three smaller questions:
What is the function (if any) of language? What are the engineering demands on a
system that must carry out such a function? And are the mechanisms of language
tailored to meet those demands? We will suggest that language show signs of
design for the communication of propositional structures over a serial channel.
3.1. An Argument for Design in Language
Humans acquire a great deal of information during their lifetimes. Since this
acquisition process occurs at a rate far exceeding that of biological evolution, it is
invaluable in dealing with causal contingencies of the environment that change
within a lifetime, and provides a decisive advantage in competition with other
species that can only defend themselves against new threats in evolutionary time
(Brandon and Hornstein, 1986; Tooby and deVore, 1987). There is an obvious
advantage in being able to acquire such information about the world second-hand:
by tapping into the vast reservoir of knowledge accumulated by some other
individual, one can avoid having to duplicate the possibly time-consuming and
dangerous trial and error process that won that knowledge. Furthermore, within a
group of interdependent, cooperating individuals, the states of other individuals are
among the most significant things in the world worth knowing about. Thus
communication of knowledge and internal states is useful to creatures who have a
lot to say and who are on speaking terms. (In Section MASTODONS, we discuss
evidence that our ancestors were such creatures.)
Human knowledge and reasoning, it has been argued, is couched in a "language of thought"
that is distinct from external languages such as English or Japanese (Fodor, 1975). The
propositions in this representational medium are relational structures whose symbols
pertain to people, objects, and events, the categories they belong to, their distribution in
space and time, and their causal relations to one another (Jackendoff, 1983; Keil, 1979).
The causal relations governing the behavior of other people are understood as involving
their beliefs and desires, which can be considered as relations between an individual and
the proposition that represents the content of that belief or desire (Fodor, 1985, 1987).
This makes the following kinds of contents as worthy of communication among humans.
We would want to be able to refer to individuals and classes, to distinguish among basic
ontological categories (things, events, places, times, manners, and so on), to talk about
events and states, distinguishing the participants in the event or state according to role
(agents, patients, goals), and to talk about the intentional states of ourselves and others.
Also, we would want the ability to express distinctions of truth value, modality (necessity,
possibility, probability, factivity), to comment on the time of an event or state including
both its distribution over time (continuous, iterative, punctate) and its overall time of
occurrence. One might also demand the ability to encode an unlimited number of
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 12
predicates, arguments, and propositions. Further, it would be useful to be able to use the
same propositional content within different speech acts; for instance, as a question, a
statement, or a command. Superimposed on all of this we might ask for an ability to focus
or to put into the background different parts of a proposition, so as to tie the speech act into
its context of previously conveyed information and patterns of knowledge of the listener.
The vocal-auditory channel has some desirable features as a medium of communication: it
has a high bandwidth, its intensity can be modulated to conceal the speaker or to cover
large distances, and it does not require light, proximity, a face-to-face orientation, or tying
up the hands. However it is essentially a serial interface, lacking the full twodimensionality needed to convey graph or tree structures and typographical devices such as
fonts, subscripts, and brackets. The basic tools of a coding scheme employing it are an
inventory of distinguishable symbols and their concatenation.
Thus grammars for spoken languages must map propositional structures onto a serial
channel, minimizing ambiguity in context, under the further constraints that the encoding
and decoding be done rapidly, by creatures with limited short-term memories, according to
a code that is shared by an entire community of potential communicants.
The fact that language is a complex system of many parts, each tailored to mapping a
characteristic kind of semantic or pragmatic function onto a characteristic kind of symbol
sequence, is so obvious in linguistic practice that it is usually not seen as worth mentioning.
Let us list some uncontroversial facts about substantive universals, the building blocks of
grammars that all theories of universal grammar posit, either as an explicit inventory or as a
consequence of somewhat more abstract mechanisms.
-Grammars are built around symbols for major lexical categories (noun, verb, adjective,
preposition) that can enter into rules specifying telltale surface distributions (e.g., verbs but
not nouns generally take unmarked direct objects), inflections, and lists of lexical items.
Together with minor categories that characteristically co-occur with the major ones (e.g.,
articles with nouns), the different categories are thus provided with the means of being
distinguished in the speech string. These distinctions are exploited to distinguish basic
ontological categories such as things, events or states, and qualities. (See, e.g., Jackendoff,
1983, 1988.)
-Major phrasal categories (noun phrase, verb phrase, etc.) start off with a major lexical
item, the "head," and allow it to be combined with specific kinds of affixes and phrases.
The resulting conglomerate is then used to refer to entities in our mental models of the
world. Thus a noun like dog does not itself describe anything but it can combine with
articles and other parts of speech to make noun phrases, such as those dogs, my dog, and
the dog that bit me, and it is these noun phrases that are used to describe things. Similarly, a
verb like hit is made into a verb phrase by marking it for tense and aspect and adding an
object, thus enabling it to describe an event. In general, words encode abstract general
categories and only by contributing to the structure of major phrasal categories can they
describe particular things, events, states, locations, and properties. This mechanism enables
the language-user to refer to an unlimited range of specific entities while possessing only a
finite number of lexical items (See, e.g., Bloom, 1989; Jackendoff, 1977).
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 13
-Phrase structure rules (e.g., "X-bar theory" or "immediate dominance rules") force
concatenation in the string to correspond to semantic connectedness in the underlying
proposition, and thus provides linear clues of underlying structure, distinguishing, for
example, Large trees grow dark berries from Dark trees grow large berries. (See, e.g.,
Gazdar, Pullum, Klein, and Sag, 1985; Jackendoff, 1977).
-Rules of linear order (e.g., "directional parameters" for ordering heads, complements, and
specifiers, or "linear precedence rules") allow the order of words within these
concatenations to distinguish among the argument positions that an entity assumes with
respect to a predicate, distinguishing Man bites dog from Dog bites man. (See, e.g., Gazdar,
et al., 1985; Travis, 1984.)
-Case affixes on nouns and adjectives can take over these functions, marking nouns
according to argument role and linking noun with predicate even when the order is
scrambled. This redundancy can free up the device of linear order, allowing it to be
exploited to convey relations of prominence and focus, which can thus mesh with the
necessarily temporal flow of attention and knowledge acquisition in the listener.
-Verb affixes signal the temporal distribution of the event that the verb refers to (aspect)
and the time of the event (tense); when separate aspect and tense affixes co-occur, they are
in a universally preferred order (aspect closer to the verb; Bybee, 1985). Given that manmade timekeeping devices play no role in species-typical human thought, some other kind
of temporal coordinates must be used, and languages employ an ingenious system that can
convey the time of an event relative to the time of the speech act itself and relative to a
third, arbitrary reference time (thus we can distinguish between John has arrived, John had
arrived (when Mary was speaking), John will have arrived (before Mary speaks), and so on;
Reichenbach, 1947). Verb affixes also typically agree with the subject and other arguments,
and thus provide another redundant mechanism that can convey predicate-argument
relations by itself (e.g., in many Native American languages such as Cherokee and Navajo)
or that can eliminate ambiguity left open by other mechanisms (distinguishing, e.g., I know
the boy and the girl who like chocolate from I know the boy and the girl who likes
chocolate).
-Auxiliaries, which occur either as verb affixes (where they are distinguished from tense
and aspect affixes by proximity to the verb) or in one of three sentence-peripheral positions
(first, second, last), convey relations that have logical scope over the entire proposition
(mirroring their peripheral position) such as truth value, modality, and illocutionary force.
(See Steele, Akmajian, Demers, Jelinek, Kitagawa, Oehrle, & Wasow, 1981).
-Languages also typically contain a small inventory of phonetically reducible morphemes -pronouns and other anaphoric elements -- that by virtue of encoding a small set of semantic
features such as gender and humanness, and being restricted in their distribution, can
convey patterns of coreference among different participants in complex relations without
the necessity of repeating lengthy definite descriptions (e.g., as in A boy showed a dog to a
girl and then he/she/it touched him/her/it/himself/herself). (See Chomsky, 1981; Wexler
and Manzini, 1984).
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 14
-Mechanisms of complementation and control govern the expression of propositions that
are arguments of other propositions, employing specific complementizer morphemes
signaling the periphery of the embedded proposition and indicating its relation to the
embedding one, and licensing the omission of repeated phrases referring to participants
playing certain combinations of roles. This allows the expression of a rich set of
propositional attitudes within a belief-desire folk psychology, such as John tried to come,
John thinks that Bill will come, John hopes for Bill to come, John convinced Bill to come,
and so on. (See Bresnan, 1982).
-In wh-movement (as in wh-questions and relative clauses) there is a tightly constrained
cooccurrence pattern between an empty element (a "trace" or "gap") and a sentenceperipheral quantifier (e.g., wh-words). The quantifier-word can be specific as to
illocutionary force (question versus modification), ontological type (time, place, purpose),
feature (animate/inanimate), and role (subject/object), and the gap can occur only in highly
constrained phrase structure configurations. The semantics of such constructions allow the
speaker to fix the reference of, or request information about, an entity by specifying its role
within any proposition. One can refer not just to any dog but to the dog that Mary sold __
to some students last year; one can ask not only for the names of just any old interesting
person but specifically Who was that woman I saw you with __? (See, e.g., Chomsky,
1981; Gazdar, Pullum, Klein, and Sag, 1985; Kaplan and Bresnan, 1982).
And this is only a partial list, focusing on sheer expressive power. One could add to it the
many syntactic constraints and devices whose structure enables them to minimize memory
load and the likelihood of pursuing local garden paths in speech comprehension (e.g.,
Berwick and Weinberg, 1984; Berwick and Wexler, 1987; Bever, 1970; Chomsky and
Lasnik, 1977; Frazier, Clifton, and Randall, 1983; Hawkins and Cutler, 1988; Kuno, 1973,
1974), or to ease the task of analysis for the child learning the language (e.g., Morgan,
1986; Pinker, 1984; Wexler and Culicover, 1980). On top of that there are the rules of
segmental phonology that smooth out arbitrary concatenations of morphemes into a
consistent sound pattern that juggles demands of ease of articulation and perceptual
distinctness; the prosodic rules that disambiguate syntax and communicate pragmatic and
illocutionary information; the articulatory programs that achieve rapid transmission rates
through parallel encoding of adjacent consonants and vowels; and on and on. Language
seems to be a fine example of "that perfection of structure and coadaptation which justly
excites our admiration" (Darwin, 1859: 26).
As we write these words, we can hear the protests: "Pangloss! Just-so stories!" Haven't we
just thought up accounts about functions post hoc after examining the structure? How do
we know that the neural mechanisms were not there for other reasons, and that once they
were there they were just put to various convenient uses by the first language users, who
then conveyed their invention to subsequent generations?
3.2. Is the Argument for Language Design a Just-So Story?
First of all, there is nothing particularly ingenious, contorted, or exotic about our
claims for substantive universals and their semantic functions. Any one of them
could have been lifted out of the pages of linguistics textbooks. It is hardly the
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 15
theory of evolution that motivates the suggestion that phrase structure rules are
useful in conveying relations of modification and predicate-argument structure.
Second, it is not necessarily illegitimate to infer both special design and adaptationist
origins on the basis of function itself. It all depends on the complexity of the function from
an engineering point of view. If someone told you that John uses X as a sunshade or a
paperweight, you would certainly be hard-pressed to guess what X is or where X came
from, because all sorts of things make good sunshades or paperweights. But if someone
told you that John uses X to display television broadcasts, it would be a very good bet that
X is a television set or is similar in structure to one, and that it was designed for that
purpose. The reason is that it would be vanishingly unlikely for something that was not
designed as a television set to display television programs; the engineering demands are
simply too complex.
This kind of reasoning is commonly applied in biology when high-tech abilities such as bat
sonar are discovered. We suggest that human language is a similar case. We are not talking
about noses holding up spectacles. Human language is a device capable of communicating
exquisitely complex and subtle messages, from convoluted soap opera plots to theories of
the origin of the universe. Even if all we knew was that humans possessed such a device,
we would expect that it would have to have rather special and unusual properties suited to
the task of mapping complex propositional structures onto a serial channel, and an
examination of grammar confirms this expectation.
Third, arguments that language is designed for communication of propositional structures
are far from logical truths. It is easy to formulate, and reject, specific alternatives. For
example, occasionally it is suggested that language evolved as a medium of internal
knowledge representation for use in the computations underlying reasoning. But while there
may be a language-like representational medium -- "the language of thought," or
"mentalese" (Fodor, 1975) -- it clearly cannot be English, Japanese, and so on. Natural
languages are hopeless for this function: they are needlessly serial, rife with ambiguity
(usually harmless in conversational contexts, but unsuited for long term knowledge
representation), complicated by alternations that are relevant only to discourse (e.g.,
topicalization), and cluttered with devices (such as phonology and much of morphology)
that make no contribution to reasoning. Similarly, the facts of grammar make it difficult to
argue that language shows design for "the expression of thought" in any sense that is
substantially distinct from "communication." If "expression" refers to the mere
externalization of thoughts, in some kind of monologue or soliloquy, it is an unexplained
fact that language contains mechanisms that presuppose the existence of a listener, such as
rules of phonology and phonetics (which map sentences onto sound patterns, enhance
confusable phonetic distinctions, disambiguate phrase structure with intonation, and so on.)
and pragmatic devices that encode conversational topic, illocutionary force, discourse
antecedents, and so on. Furthermore people do not express their thoughts in an arbitrary
private language (which would be sufficient for pure "expression"), but have complex
learning mechanisms that acquire a language highly similar in almost every detail to those
of other speakers in the community.
Another example of of the empirical nature of specific arguments for language design
appears when we examine the specific expressive abilities that are designed into language.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 16
They turn out to constitute a well-defined set, and do not simply correspond to every kind
of information that humans are interested in communicating. So although we may have
some a priori intuitions regarding useful expressive capacities of grammar, the matter is
ultimately empirical (see, e.g., Jackendoff, 1983, 1990; Pinker, 1989b; Talmy, 1983, 1988),
and such research yields results that are specific enough to show that not just any intuition
is satisfied. Grammar is a notoriously poor medium for conveying subtle patterns of
emotion, for example, and facial expressions and tones of voice are more informative
(Ekman and Friesen, 1975; Etcoff, 1986). Although grammars provide devices for
conveying rough topological information such as connectivity, contact, and containment,
and coarse metric contrasts such as near/far or flat/globular, they are of very little help in
conveying precise Euclidean relations: a picture is worth a thousand words. Furthermore,
human grammar clearly lacks devices specifically dedicated to expressing any of the kinds
of messages that characterize the vocal communication systems of cetaceans, birds, or
nonhuman primates, such as announcements of individual identity, predator warnings, and
claims of territory.
Finally, Williams (1966) suggests that convergent evolution, resemblance to man-made
artifacts, and direct assessments of engineering efficiency are good sources of evidence for
adaptation. Of course in the case of human language these tests are difficult in practice:
significant convergent evolution has not occurred, no one has ever invented a system that
duplicates its function (except for systems that are obviously parasitic on natural languages
such as Esperanto or signed English), and most forms of experimental intervention would
be unethical. Nonetheless, some tests are possible in principle, and this is enough to refute
reflexive accusations of circularity.
For example, even the artificial languages that are focussed on very narrow domains of
content and that are not meant to be used in a natural on-line manner by people, such as
computer languages or symbolic logic, show certain obvious parallels with aspects of
human grammar. They have needed means of distinguishing types of symbols, predicate
argument relations, embedding, scope, quantification, and truth relations, and solve these
problems with formal syntactic systems that specify arbitrary patterns of hierarchical
concatenation, relative linear order, fixed positions within strings, and closed classes of
privileged symbols. Of course there are vast dissimilarities but the mere fact that terms like
"language," "syntax," "predicate," "argument," and "statement" have clear meanings when
applied to artificial systems, with no confusion or qualification, suggests that there are
nonaccidental parallels that are reminiscent of the talk of diaphragms and lenses when
applied to cameras and eyes. As for experimental investigation, in principle one could
define sets of artificial grammars with and without one of the mechanisms in question, or
with variations of it. The grammars would be provided or taught to pairs of communicators
-- formal automata, computer simulations, or college sophomores acting in conscious
problem-solving mode -- who would be required to convey specific messages under
different conditions of speed, noise, or memory limitations. The proportion of information
successfully communicated would be assessed and examined as a function of the presence
and version of the grammatical mechanism, and of the different conditions putatively
relevant to the function in question.
3.3. Language Design and Language Diversity
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 17
A more serious challenge to the claim that grammars show evidence of good
design may come from the diversity of human languages (Maratsos, 1989).
Grammatical devices and expressive functions do not pair up in one-to-one
fashion. For example, some languages use word order to convey who did what to
whom; others use case or agreement for this purpose and reserve the use of word
order to distinguish topic from comment, or do not systematically exploit word order
at all. How can one say that the mental devices governing word order evolved
under selection pressure for expressing grammatical relations if many languages
do not use them for that purpose? Linguistic diversity would seem to imply that
grammatical devices are very general-purpose tools. And a general-purpose tool
would surely have a very generalized structure, and thus could be a spandrel
rather than an adapted machine. We begin by answering the immediate objection
that the existence of diversity, for whatever reason, invalidates arguments for
universal language design; at the end of the section we offer some speculations as
to why there should be more than one language to begin with.
First of all, the evolution of structures that serve not one but a small number of definite
functions, perhaps to different extents in different environments, is common in biology
(Mayr, 1982). Indeed, though grammatical devices are put to different uses in different
languages, the possible pairings are very circumscribed. No language uses noun affixes to
express tense or elements with the syntactic privileges of auxiliaries to express the shape of
the direct object. Such universal constraints on structure and function are abundantly
documented in surveys of the languages of the world (e.g., Bybee, 1985; Comrie, 1981;
Greenberg, 1966; Greenberg, Ferguson, and Moravcsik, 1978; Hawkins, 1988; Keenan,
1987; and Shopen, 1985). Moreover language universals are visible in language history,
where changes tend to fall into a restricted set of patterns, many involving the introduction
of grammatical devices obeying characteristic constraints (Kiparsky, 1976; Wang,
1976).(Note 3)
But accounting for the evolution of a language faculty permitting restricted variation is only
important on the most pessimistic of views. Even a smidgin of grammatical analysis reveals
that surface diversity is often a manifestation of minor differences in the underlying mental
grammars. Consider some of the supposedly radical typological differences between
English and other languages. English is a rigid word-order language; in the Australian
language Warlpiri the words from different logical units can be thoroughly scrambled and
case markers are used to convey grammatical relations and noun modification. Many
Native American languages, such as Cherokee, use few noun phrases within clauses at all,
and express grammatical relations by sticking strings of agreement affixes onto the verb,
each identifying an argument by a set of features such as humanness or shape. Whereas
"accusative" languages like English collapse subjects of transitive and intransitive
sentences, "ergative" languages collapse objects of transitives with subjects of intransitives.
Whereas English sentences are built around obligatory subjects, languages like Chinese are
oriented around a position reserved for the discourse topic.
However, these variations almost certainly correspond to differences in the extent to which
the same specific set of mental devices is put to use, but not to differences in the kinds of
devices that are put to use. English has free constituent order in strings of prepositional
phrases (The package was sent from Chicago to Boston by Mary; The package was sent by
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 18
Mary to Boston from Chicago, and so on). English has case, both in pronouns and in the
genitive marker spelled 's. It expresses information about arguments in verb affixes in the
agreement marker -s. Ergativity can be seen in verb alternations like John broke the glass
and The glass broke. There is even a kind of topic position: As for fish, I like salmon.
Conversely, Warlpiri is not without phrasal syntax. Auxiliaries go in second position (not
unlike English, German, and many other languages). The constituents of a noun phrase
must be contiguous if they are not case-marked; the constituents of a finite clause must be
contiguous if the sentence contains more than one. Pinker (1984) outlines a theory of
language acquisition in which the same innate learning mechanisms are put to use to
different extents in children acquiring "radically" different languages.
When one looks at more abstract linguistic analyses, the underlying unity of natural
languages is even more apparent. Chomsky has quipped that anything you find in one
language can also be found in every other language, perhaps at a more abstract level of
representation, and this claim can be justified without resorting to Procrustean measures. In
many versions of his Government-Binding theory (1981), all noun phrases must be case
marked; even those that receive no overt case-marking are assigned "abstract" case by an
adjacent verb, preposition, or tense element. The basic order of major phrases is determined
by the value of a language-varying parameter specifying the direction in which case
assignment may be executed. So in a language like Latin, the noun phrases are marked with
morphological case (and can appear in any position), while in a language like English, they
are not so marked, and must be adjacent to a case-assigner such as a verb. Thus overt case
marking in one language and word order in another are unified as manifestations of a single
grammatical module. And the module has a well-specified function: in the terminology of
the theory, it makes noun phrases "visible" for the assignment of thematic roles such as
agent, goal, or location. Moreover, word order itself is not a unified phenomenon. Often
when languages "use word order for pragmatic purposes," they are exploiting an underlying
grammatical subsystem, such as stylistic rules, that has very different properties from that
governing the relative order of noun phrases and their case-assigners.
Why is there more than one language at all? Here we can only offer the most tentative of
speculations. For sound-meaning pairings within the lexicon, there are two considerations.
First, one might suppose that speakers need a learning mechanism for labels for cultural
innovations, such as screwdriver. Such a learning device is then sufficient for all
vocabulary items. Second, it may be difficult to evolve a huge innate code. Each of tens of
thousands of sound-meaning correspondences would have to be synchronized across
speakers, but few words could have the nonarbitrary antecedents that would have been
needed to get the standardization process started (i.e., analogous to the way bared fangs in
preparation for biting evolved into the facial expression for anger.) Furthermore the size of
such a code would tax the time available to evolve and maintain it in the genome in the face
of random perturbations from sexual recombination and other stochastic genetic processes
(Williams, 1966; Tooby and Cosmides, 1989). Once a mechanism for learning soundmeaning pairs is in place, the information for acquiring any particular pair, such as dog for
dogs, is readily available from the speech of the community. Thus the genome can store the
vocabulary in the environment, as Tooby and Cosmides (1989) have put it.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 19
For other aspects of grammar, one might get more insight by inverting the perspective.
Instead of positing that there are multiple languages, leading to the evolution of a
mechanism to learn the differences among them, one might posit that there is a learning
mechanism, leading to the development of multiple languages. That is, some aspects of
grammar might be easily learnable from environmental inputs by cognitive processes that
may have been in existence prior to the evolution of grammar, for example, the relative
order of a pair of sequenced elements within a bounded unit. For these aspects there was no
need to evolve a fixed value, and they are free to vary across communities of speakers. In
Section SHARED we discuss a simulation of evolution by Hinton and Nowlan (1987) that
behaves in a way that is consistent with this conjecture.
3.4. Language Design and Arbitrariness
Piattelli-Palmarini (1989) presents a different kind of argument: grammar is not
completely predictable as an adaptation to communication, therefore it lacks
design and did not evolve by selection. He writes, "Survival criteria, the need to
communicate and plan concerted action, cannot account for our specific linguistic
nature. Adaptation cannot even begin to explain any of these phenomena."
Frequently cited examples of arbitrary phenomena in language include constraints
on movement (such as subjacency), irregular morphology, and lexical differences
in predicate-argument structure. For instance, it is acceptable to say Who did John
see Mary with?, but not Who did John see Mary and?; John broke the glass but not
John breaked the glass;John filled the glass with milk, but not John poured the
glass with milk. The arguments that language could not be an adaptation take two
forms: (i) language could be better than it is, and (ii) language could be different
than it is. We show that neither form of the argument is valid, and that the facts that
it invokes are perfectly consistent with language being an adaptation and offer not
the slightest support to any specific alternative.
3.4.1. Inherent Tradeoffs
In their crudest form, arguments about the putative functionlessness of grammar
run as follows: "I bet you can't tell me a function for Constraint X; therefore
language is a spandrel." But even if it could be shown that one part of language
had no function, that would not mean that all parts of language had no function.
Recall from Section 2.2 that many organs contain modified spandrels but this does
not mean that natural selection did not assemble or shape the organ. Worse,
Constraint X may not be a genuine part of the language faculty but just a
description of one aspect of it, an epiphenomenal spandrel. No adaptive organ can
be adaptive in every aspect, because there are as many aspects of an organ as
there are ways of describing it. The recent history of linguistics provides numerous
examples where a newly-discovered constraint is first proposed as an explicit
statement listed as part of a grammar, but is then shown to be a deductive
consequence of a far more wide-ranging principle (see, e.g., Chomsky, 1981;
Freidin, 1978.) For example, the ungrammaticality of sentences like John to have
won is surprising, once attributed to a filter specifically ruling out [NP-to-VP]
sequences, is now seen as a consequence of the Case Filter. Although one might
legitimately wonder what good "*[NP-to-VP]" is doing in a grammar, one could
hardly dispense with something like the Case Filter.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 20
Since the mere appearance of some nonoptimal feature is inconclusive, we must examine
specific explanations for why the feature exists. In the case of the nonselectionist position
espoused by Piattelli-Palmarini, there is none: not a hint of how any specific aspect of
grammar might be explained, even in principle, as a specific consequence of some
developmental process or genetic mechanism or constraint on possible brain structure. The
position gains all its support from the supposed lack of an adaptive explanation. In fact, we
will show that there is such an explanation, well-motivated both within evolutionary theory
and within linguistics, so the support disappears.
The idea that natural selection aspires toward perfection has long been discredited within
evolutionary theory (Williams, 1966). As Maynard Smith (1984: 290) has put it, "If there
were no constraints on what is possible, the best phenotype would live forever, would be
impregnable to predators, would lay eggs at an infinite rate, and so on." Tradeoffs among
conflicting adaptive goals are a ubiquitous limitation on optimality in the design of
organisms. It may be adaptive for a male bird to advertise his health to females with gaudy
plumage or a long tail, but not to the extent that predators are attracted or flight is
impossible.
Tradeoffs of utility within language are also unavoidable (Bolinger, 1980; Slobin, 1977).
For example, there is a conflict of interest between speaker and hearer. Speakers want to
minimize articulatory effort and hence tend towards brevity and phonological reduction.
Hearers want to minimize the effort of understanding and hence desire explicitness and
clarity. This conflict of interest is inherent to the communication process and operates at
many levels. Editors badger authors into expanding elliptical passages; parsimonious
headline writers unwittingly produce Squad Helps Dog Bite Victim and Stud Tires Out.
Similarly there is a conflict of interest between speaker and learner. A large vocabulary
allows for concise and precise expression. But it is only useful if every potential listener has
had the opportunity to learn each item. Again, this tradeoff is inherent to communication;
one man's jargon term is another's mot juste.
Clearly, any shared system of communication is going to have to adopt a code that is a
compromise among these demands, and so will appear to be arbitrary from the point of
view of any one criterion. There is always a large range of solutions to the combined
demands of communication that reach slightly different equilibrium points in this
multidimensional space. Slobin (1977) points out that the Serbo-Croatian inflectional
system is "a classic Indo-European synthetic muddle," suffixing each noun with a single
affix from a paradigm full of irregularity, homophony, and zero-morphemes. As a result the
system is perfected late and with considerable difficulty. In contrast the Turkish inflectional
system is semantically transparent, with strings of clearly demarcated regular suffixes, and
is mastered by the age of two. When it comes to production by an adult who has
overlearned the system, however, Serbo-Croatian does have an advantage in minimizing
the sheer number of syllables that must be articulated. Furthermore, Slobin points out that
such tradeoffs can be documented in studies of historical change and borrowing. For
example changes that serve to enhance brevity will proceed until comprehension becomes
impaired, at which point new affixes or distinctions are introduced to restore the balance
(see also Samuels, 1972). A given feature of language may be arbitrary in the sense that
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 21
there are alternative solutions that are better from the standpoint of some single criterion.
But this does not mean that it is good for nothing at all!
Subjacency -- the prohibition against dependencies between a gap and its antecedent that
spans certain combinations of phrasal nodes -- is a classic example of an arbitrary
constraint (see Freidin & Quicoli, 1989). In English you can say What does he believe they
claimed that I said? but not the semantically parallel *What does he believe the claim that I
said?. One might ask why languages behave this way. Why not allow extraction anywhere,
or nowhere? The constraint may exist because parsing sentences with gaps is a notoriously
difficult problem and a system that has to be prepared for the possibility of inaudible
elements anywhere in the sentence is in danger of bogging down by positing them
everywhere. Subjacency has been held to assist parsing because it cuts down on the set of
structures that the parser has to keep track of when finding gaps (Berwick and Weinberg,
1984). This bonus to listeners is often a hindrance to speakers, who struggle with
resumptive pronouns in clumsy sentences such as That's the guy that you heard the rumor
about his wife leaving him. There is nothing "necessary" about the precise English version
of the constraint or about the small sample of alternatives allowed within natural language.
But by settling in on a particular subset of the range of possible compromises between the
demands of expressiveness and parsability, the evolutionary process may have converged
on a satisfactory set of solutions to one problem within language processing.
3.4.2. Parity in Communications Protocols
The fact that one can conceive of a biological system being different than it is says
nothing about whether it is an adaptation (see Mayr, 1983). No one would argue
that selection was not the key organizing force in the evolution of the vertebrate
eye just because the compound eyes of arthropods are different. Similarly, pointing
out that a hypothetical Martian language could do passivization differently is
inconclusive. We must ask how well-supported specific explanations are.
In the case of features of human language structure that could have been different, again
Piattelli-Palmarini presents no explanations at all and relies entirely on the putative inability
of natural selection to provide any sort of motivated account. But in fact there is such an
account: the nature of language makes arbitrariness of grammar itself part of the adaptive
solution of effective communication in principle.
Any communicative system requires a coding protocol which can be arbitrary as long as it
is shared. Liberman and Mattingly (1989) call this the requirement of parity, and we can
illustrate it with the (coincidentally-named) "parity" settings in electronic communication
protocols. There is nothing particularly logical about setting your printer's serial interface to
the "even," as opposed to the "odd," parity setting. Nor is there any motivation to set your
computer to odd as opposed to even parity. But there is every reason to set the computer
and printer to the same parity, whatever it is, because if you don't, they cannot
communicate. Indeed, standardization itself is far more important than any other adaptive
feature possessed by one party. Many personal computer manufacturers in the 1980s
boasted of the superior engineering and design of their product compared to the IBM PC.
But when these machines were not IBM-compatible, the results are well-known.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 22
In the evolution of the language faculty many "arbitrary" constraints may have been
selected simply because they defined parts of a standardized communicative code in the
brains of some critical mass of speakers. Piattelli-Palmarini may be correct in claiming that
there is nothing adaptive about forming yes-no questions by inverting the subject and
auxiliary as opposed to reversing the order of words in the sentence. But given that
language must do one or the other, it is highly adaptive for each member of a community of
speakers to be forced to learn to do it the same way as all the other members. To be sure,
some combination of historical accidents, epiphenomena of other cognitive processes, and
neurodevelopmental constraints must have played a large role in the breaking of symmetry
that was needed to get the fixation process running away in one direction or another. But it
still must have been selection that resulted in the convention then becoming innately
entrenched.
The requirement of parity operates at all levels of a communications protocol. Within
individual languages the utility of arbitrary but shared features is most obvious in the
choice of individual words: there is no reason for you to call a dog dog rather than cat
except for the fact that everyone else is doing it, but that is reason enough. Saussure (1959)
called this inherent feature of language "l'arbitraire du signe," and Hurford (1989), using
evolutionary game theory, demonstrates the evolutionary stability of such a "Saussurean"
strategy whereby each learner uses the same arbitrary signs in production that it uses in
comprehension (i.e., that other speakers use in production). More generally, these
considerations suggest that a preference for arbitrariness is built into the language
acquisition device at two levels. It only hypothesizes rules that fall within the (possibly
arbitrary) set defined by universal grammar, and within that set, it tries to choose rules that
match those used by the community, whatever they are.
The benefits of a learning mechanism designed to assess and adopt the prevailing parity
settings become especially clear when we consider alternatives, such as trying to get each
speaker to converge on the same standard by endogenously applying some rationale to
predict form from meaning. There are many possible rationales for any form-meaning
pairing, and that is exactly the problem -- different rationales can impress different
speakers, or the same speakers on different occasions, to different degrees. But such
differences in cognitive style, personal history, or momentary interests must be set aside if
people are to communicate. As mentioned, no grammatical device can simultaneously
optimize the demands of talkers and hearers, but it will not do to talk in Serbo-Croatian and
demand that one's listeners reply in Turkish. Furthermore, whenever cognition is flexible
enough to construe a situation in more than one way, no simple correspondence between
syntax and semantics can be used predictively by a community of speakers to "deduce" the
most "logical" grammatical structure. For example, there is a simple and universal principle
dictating that the surface direct object of a causative verb refers to an entity that is
"affected" by the action. But the principle by itself is unusable. When a girl puts boxes in
baskets she is literally affecting both: the boxes are changing location, and the baskets are
changing state from empty to full. One would not want one perceiver interested in the
boxes to say that she is filling boxes while another interested in the baskets to describe the
same event as filling baskets; no one would know what went where. However by letting
different verbs idiosyncratically select different kinds of entities as "affected" (e.g., place
the box/*basket versus fill the basket/*box), and forcing learners to respect the verbs'
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 23
wishes, grammar can allow speakers to specify different kinds of entities as affected by
putting them in the direct object position of different verbs, with minimal ambiguity.
Presumably this is why different verbs have different arbitrary syntactic privileges (Pinker,
1989b), a phenomenon that Piattelli-Palmarini (1989) describes at length. Even iconicity
and onomatopoeia are in the eye and ear of the beholder. The ASL sign for "tree" resembles
the motion of a tree waving in the wind, but in Chinese Sign Language it is the motion of
sketching the trunk (Newport & Meier, 1985). In the United States, pigs go "oink"; in
Japan, they go "boo-boo."
3.4.3. Arbitrariness and the Relation Between Language Evolution and Language
Acquisition
The need for arbitrariness has profound consequences for understanding the role of
communicative function in language acquisition and language evolution. Many
psychologists and artificial intelligence researchers have suggested that the structure of
grammar is simply the solution that every child arrives at in solving the problem of how to
communicate with others. Skinner's reinforcement theory is the strongest version of this
hypothesis (Skinner, 1957), but versions that avoid his behaviorism and rely instead on
general cognitive problem-solving abilities have always been popular within psychology.
Both Skinner and cognitive theorists such as Bates et al. (1989) explicitly draw parallels
between the role of function in learning and evolution. Chomsky and many other linguists
and psycholinguists have argued against functionalism in ontogeny, showing that many
aspects of grammar cannot be reduced to being the optimal solution to a communicative
problem; rather, human grammar has a universal idiosyncratic logic of its own. More
generally, Chomsky has emphasized that people's use of language does not tightly serve
utilitarian goals of communication but is an autonomous competence to express thought
(see, e.g., Chomsky, 1975). If communicative function does not shape language in the
individual, one might conclude, it probably did not shape language in the species.
We suggest that the analogy that underpins this debate is misleading. It is not just that
learning and evolution need not follow identical laws, selectionist or otherwise. (For
example, as Chomsky himself has stressed, the issue never even comes up in clearer cases
like vision, where nobody suggests that all infants' visual development is related to their
desire to see or that visual systems develop with random variations that are selected by
virtue of their ability to attain the child's goals.) In the case of language the arguments of
section 3.4 suggest that language evolution and language acquisition not only can differ but
that they must differ. Evolution has had a wide variety of equivalent communicative
standards to choose from; there is no reason for it to have favored the class of languages
that includes Apache and Yiddish, but not Old High Martian or Early Vulcan. But this
flexibility has been used up by the time a child is born; the species and the language
community have already made their choices. The child cannot learn just any useful
communicative system; nor can he or she learn just any natural language. He or she is stuck
with having to learn the particular kind of language the species eventually converged upon
and the particular variety the community has chosen. Whatever rationales may have
influenced these choices are buried in history and cannot be recapitulated in development.
Moreover, any code as complex and precise as a grammar for a natural language will not
wear its protocol on its sleeve. No mortal computer user can induce an entire
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 24
communications protocol or programming language from examples; that's why we have
manuals. This is because any particular instance of the use of such a protocol is a unique
event accompanied by a huge set of idiosyncratic circumstances, some relevant to how the
code must be used, most irrelevant, and there is no way of deciding which is which. For the
child, any sentence or set of sentences is compatible with a wide variety of very different
grammars, only one of them correct (Chomsky, 1965, 1975, 1980, 1981; Pinker, 1979,
1984; Wexler and Culicover, 1980). For example, without prior constraints, it would be
natural to generalize from input sentences like Who did you see her with? to *Who did you
see her and?, from teethmarks to *clawsmarks, from You better be good to *Better you be
good?. The child has no manual to consult, and presumably that is why he or she needs
innate constraints.
So we see a reason why functionalist theories of the evolution of language can be true while
functionalist theories of the acquisition of language can be false. From the very start of
language acquisition, children obey grammatical constraints that afford them no immediate
communicative advantage. To take just one example, 1- and 2-year-olds acquiring English
obey a formal constraint on phrase structure configurations concerning the distinction
between lexical categories and phrasal categories and as a result avoid placing determiners
and adjectives before pronouns and proper names. They will use phrases like big dog to
express the belief that a particular dog is big, but they will never use phrases like big Fred
or big he to express the belief that a particular person is big (Bloom, in press). Children
respect this constraint despite the limits it puts on their expressive range.
Furthermore, despite unsupported suggestions to the contrary among developmental
psychologists, many strides in language development afford the child no locally-discernible
increment in communicative ability (Maratsos, 1983, 1989). When children say breaked
and comed, they are using a system that is far simpler and more logical than the adult
combination of a regular rule and 150 irregular memorized exceptions. Such errors do not
reliably elicit parental corrections or other conversational feedback (Brown and Hanlon,
1970; Morgan and Travis, in press). There is no deficit in comprehensibility; the meaning
of comed is perfectly clear. In fact the child's system has greater expressive power that the
adult's. When children say hitted and cutted, they are distinguishing between past and
nonpast forms in a manner that is unavailable to adults, who must use hit and cut across the
board. Why do children eventually abandon this simple, logical, expressive system? They
must be programmed so that the mere requirement of conformity to the adult code, as subtle
and arbitrary as it is, wins over other desiderata.
The requirement that a communicative code have an innate arbitrary foundation ("universal
grammar," in the case of humans) may have analogues elsewhere in biology. Mayr (1982:
612) notes that
Behavior that serves as communication, for instance courtship behavior, must be
stereotyped in order not to be misunderstood. The genetic program controlling such
behavior must be "closed," that is, it must be reasonably resistant to any changes during the
individual life cycle. Other behaviors, for instance those that control the choice of food or
habitat, must have a certain amount of flexibility in order to permit the incorporation of
new experiences; such behaviors must be controlled by an "open" program.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 25
In sum, the requirement for standardization of communication protocols dictates that it is
better for nature to build a language acquisition device that picks up the code of the ambient
language than one that invents a code that is useful from a child's eye view. Acquiring such
a code from examples is no mean feat, and so many grammatical principles and constraints
must be hardwired into the device. Thus even if the functions of grammatical devices play
an important role in evolution, they may play no role in acquisition.
4. Arguments for Language Being a Spandrel
Given that the criteria for being an adaptation appear to be satisfied in the case of
language, we can examine the strength of the competing explanations that
language is a spandrel suggested by Gould, Chomsky, and Piattelli-Palmarini.
4.1. The Mind as a Multipurpose Learning Device
The main motivation for Gould's specific suggestion that language is a spandrel is
his frequently-stated position that the mind is a single general-purpose computer.
For example, as part of a critique of a theory of the origin of language, Gould
(1979: 386) writes:
I don't doubt for a moment that the brain's enlargement in human evolution had an adaptive
basis mediated by selection. But I would be more than mildly surprised if many of the
specific things it now can do are the product of direct selection "for" that particular
behavior. Once you build a complex machine, it can perform so many unanticipated tasks.
Build a computer "for" processing monthly checks at the plant, and it can also perform
factor analyses on human skeletal measures, play Rogerian analyst, and whip anyone's ass
(or at least tie them perpetually) in tic-tac-toe.
The analogy is somewhat misleading. It is just not true that you can take a computer that
processes monthly checks and use it to play Rogerian analyst; someone has to reprogram it
first. Language learning is not programming: parents provide their children with sentences
of English, not rules of English. We suggest that natural selection was the programmer.
The analogy could be modified by imagining some machine equipped with a single
program that can learn from examples to calculate monthly checks, perform factor analyses,
and play Rogerian analyst, all without explicit programming. Such a device does not now
exist in artificial intelligence and it is unlikely to exist in biological intelligence. There is no
psychologically realistic multipurpose learning program that can acquire language as a
special case, because the kinds of generalizations that must be made to acquire a grammar
are at cross-purposes with those that are useful in acquiring other systems of knowledge
from examples (Chomsky, 1982; Pinker, 1979, 1984; Wexler and Culicover, 1980). The
gross facts about the dissociability of language and other learned cultural systems, listed in
the first paragraph of this paper, also belie the suggestion that language is a spandrel of any
general cognitive learning ability.
4.2. Constraints on Possible Forms
The theory that the mind is an all-purpose learning device is of course anathema to
Chomsky (and to Piattelli-Palmarini), making it a puzzle that they should find
themselves in general agreement with Gould. Recently, Gould (1989) has
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 26
described some common ground. Chomsky, he suggests, is in the Continental
tradition of trying to explain evolution by structural laws constraining possible
organic forms. For example, Chomsky writes:
In studying the evolution of mind, we cannot guess to what extent there are physically
possible alternatives to, say, transformational generative grammar, for an organism meeting
certain other physical conditions characteristic of humans. Conceivably, there are none -- or
very few -- in which case talk about evolution of the language capacity is beside the point.
(1972: 97-98).
These skills [e.g., learning a grammar] may well have arisen as a concomitant of structural
properties of the brain that developed for other reasons. Suppose that there was selection
for bigger brains, more cortical surface, hemispheric specialization for analytic processing,
or many other structural properties that can be imagined. The brain that evolved might well
have all sorts of special properties that are not individually selected; there would be no
miracle in this, but only the normal workings of evolution. We have no idea, at present,
how
10 physical laws apply when 10 neurons are placed in an object the size of a basketball,
under the special conditions that arose during human evolution. (1982: 321)
In this regard [the evolution of infinite digital systems], speculations about natural selection
are no more plausible than many others; perhaps these are simply emergent physical
properties of a brain that reaches a certain level of complexity under the specific conditions
of human evolution. (1988b: 22 in ms.)
Although Chomsky does not literally argue for any specific evolutionary hypothesis, he
repeatedly urges us to consider "physical laws" as possible alternatives to natural selection.
But it is not easy to see exactly what we should be considering. It is certainly true that
natural selection cannot explain all aspects of the evolution of language. But is there any
reason to believe that there are as-yet undiscovered theorems of physics that can account
for the intricate design of natural language? Of course human brains obey the laws of
physics, and always did, but that does not mean that their specific structure can be
explained by such laws.
More plausibly, we might look to constraints on the possible neural basis for language and
its epigenetic growth. But neural tissue is wired up by developmental processes that act in
similar ways all over the cortex and to a lesser degree across the animal kingdom (Dodd
and Jessell, 1988; Harrelson and Goodman, 1988). In different organisms it has evolved the
ability to perform the computations necessary for pollen-source communication, celestial
navigation, Doppler-shift echolocation, stereopsis, controlled flight, dam-building, sound
mimicry, and face recognition. The space of physically possible neural systems thus can't
be all that small, as far as specific computational abilities are concerned. And it is most
unlikely that laws acting at the level of substrate adhesion molecules and synaptic
competition, when their effects are projected upward through many levels of scale and
hierarchical organization, would automatically result in systems that accomplish interesting
engineering tasks in a world of medium-sized objects.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 27
Changes in brain quantity could lead to changes in brain quality. But mere largeness of
brain is neither a necessary nor a sufficient condition for language, as Lenneberg's (1967)
studies of nanencephaly and craniometric studies of individual variation have shown. Nor is
there reason to think that if you simply pile more and more neurons into a circuit or more
and more circuits into a brain that computationally interesting abilities would just emerge.
It seems more likely that you would end up with a very big random pattern generator.
Neural network modeling efforts have suggested that complex computational abilities
require either extrinsically imposed design or numerous richly structured inputs during
learning or both (Pinker & Prince, 1988; Lachter & Bever, 1988), any of which would be
inconsistent with Chomsky's suggestions.
Finally, there may be direct evidence against the speculation that language is a necessary
physical consequence of how human brains can grow. Gopnik (1990a, b) describes a
syndrome of developmental dysphasia whose sufferers lack control of morphological
features such as number, gender, tense, and case. Otherwise they are intellectually normal.
One 10-year-old boy earned the top grade in his mathematics class and is a respectable
computer programmer. This shows that a human brain lacking components of grammar,
perhaps even a brain with the capacity of discrete infinity, is physically and
neurodevelopmentally possible.
In sum, there is no support for the hypothesis that language emerges from physical laws
acting in unknown ways in a large brain. While there are no doubt aspects of the system
that can only be explained by historical, developmental, or random processes, the most
likely explanation for the complex structure of the language faculty is that it is a design
imposed on neural circuitry as a response to evolutionary pressures.
5. The Process of Language Evolution
For universal grammar to have evolved by Darwinian natural selection, it is not enough that
it be useful in some general sense. There must have been genetic variation among
individuals in their grammatical competence. There must have been a series of steps
leading from no language at all to language as we now find it, each step small enough to
have been produced by a random mutation or recombination, and each intermediate
grammar useful to its possessor. Every detail of grammatical competence that we wish to
ascribe to selection must have conferred a reproductive advantage on its speakers, and this
advantage must be large enough to have become fixed in the ancestral population. And
there must be enough evolutionary time and genomic space separating our species from
nonlinguistic primate ancestors.
There are no conclusive data on any of these issues. However this has not prevented various
people from claiming that each of the necessary postulates is false! We argue that what we
do know from the biology of language and evolution makes each of the postulates quite
plausible.
5.1. Genetic Variation
Lieberman (1984, 1989) claims that the Chomskyan universal grammar could not
have evolved. He writes:
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 28
The premises that underlie current "nativist" linguistic theory
... are out of touch with modern biology. Ernst Mayr (1982), in his definitive work, The
Growth of Biological Thought, discusses these basic principles that must structure any
biologically meaningful nativist theory. ... [one of the principles is:] Essentialistic thinking
(e.g., characterizing human linguistic ability in terms of a uniform hypothetical universal
grammar) is inappropriate for describing the biological endowment of living organisms.
(1989: 203-205)
A true nativist theory must accommodate genetic variation. A detailed genetically
transmitted universal grammar that is identical for every human on the planet is outside the
range of biological plausibility. (1989: 223)
This is part of Lieberman's argument that syntax is acquired by general-purpose learning
abilities, not by a dedicated module or set of modules. But the passages quoted above
contain a variety of misunderstandings and distortions. Chomskian linguistics is the
antithesis of the kind of essentialism that Mayr decries. It treats such disembodied
interindividual entities as "The English Language" as unreal epiphenomena. The only
scientifically genuine entities are individual grammars situated in the heads of individual
speakers (see Chomsky, 1986, for extended discussion). True, grammars for particular
languages, and universal grammar, are often provisionally idealized as a single kind of
system. But this is commonplace in systems-level physiology and anatomy; for example the
structure of the human eye is always described as if all individuals shared it and individual
variation and pathology are discussed as deviations from a norm. This is because natural
selection, while feeding on variation, uses it up (Ridley, 1986; Sober, 1984). In adaptively
complex structures in particular, the variation we see does not consist of qualitative
differences in basic design, and this surely applies to complex mental structures as well
(Tooby and Cosmides, 1989).
Also, contrary to what Lieberman implies, there does exist variation in grammatical ability.
Within the range that we would call "normal" we all know some individuals who habitually
use tangled syntax and others who speak with elegance, some who are linguistically
creative and others who lean on cliches, some who are fastidious conformists and others
who bend and stretch the language in various ways. At least some of this variation is
probably related to the strength or accessibility of different grammatical subsystems, and at
least some, we suspect, is genetic, the kind of thing that would be shared by identical twins
reared apart. More specifically, Bever, Carrithers, Cowart, and Townsend (1989) have
extensive experimental data showing that right-handers with a family history of lefthandedness show less reliance on syntactic analysis and more reliance on lexical
association than do people without such a genetic background.
Moreover, beyond the "normal" range there are documented genetically-transmitted
syndromes of grammatical deficits. Lenneberg (1967) notes that specific language
disability is a dominant partially sex-linked trait with almost complete penetrance (see also
Ludlow and Cooper, 1983, for a literature review). More strikingly, Gopnik (1990b) has
found a familial selective deficit in the use of morphological features (gender, number,
tense, etc.) that acts as if it is controlled by a dominant gene.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 29
This does not mean that we should easily find cases of inherited subjacency deficiency or
anaphor blindness. Pleiotropy -- single gene changes that cause apparently unrelated
phenotypic effects -- is ubiquitous, so there is no reason to think that every aspect of
grammar that has a genetic basis must be controlled by a single gene. Having a right hand
has a genetic basis but genetic deficits do not lead to babies being born with exactly one
hand missing. Moreover, even if there was a pure lack of some grammatical device among
some people, it may not be easily discovered without intensive analysis of the person's
perceptions of carefully constructed linguistic examples. Different grammatical subsystems
can generate superficially similar constructions and a hypothetical victim of a deficit may
compensate in ways that would be difficult to detect. Indeed cases of divergent underlying
analyses of a single construction are frequent causes of historical change.
5.2. Intermediate Steps
Some people have doubted that an evolutionary sequence of increasingly complex
and specialized universal grammars is possible. The intermediate links, it has been
suggested, would not have been viable communication systems. These arguments
fall into three classes.
5.2.1. Nonshared Innovations.
Geschwind (1980), among others, has wondered how a hypothetical "beneficial"
grammatical mutation could really have benefited its possessor, given that none of
the person's less evolved compatriots could have understood him or her. One
possible answer is that any such mutation is likely to be shared by individuals who
are genetically-related. Since much communication is among kin, a linguistic
mutant will be understood by some of his or her relatives and the resulting
enhancements in information sharing will benefit each one of them relative to
others who are not related.
But we think there is a more general answer. Comprehension abilities do not have to be in
perfect synchrony with production abilities. Comprehension can use cognitive heuristics
based on probable events to decode word sequences even in the absence of grammatical
knowledge. Ungrammatical strings like skid crash hospital are quite understandable, and
we find we can do a reasonably good job understanding Italian newspaper stories based on
a few cognates and general expectancies. At the same time grammatical sophistication in
such sources does not go unappreciated. We are unable to duplicate Shakespeare's complex
early Modern English but we can appreciate the subtleties of his expressions. When some
individuals are making important distinctions that can be decoded with cognitive effort, it
could set up a pressure for the evolution of neural mechanisms that would make this
decoding process become increasingly automatic, unconscious, and undistracted by
irrelevant aspects of world knowledge. These are some of the hallmarks of an innate
grammatical "module" (Fodor, 1983). The process whereby environmentally-induced
responses set up selection pressures for such responses to become innate, triggering
conventional Darwinian evolution that superficially mimics a Lamarckian sequence, is
sometimes known as the Baldwin Effect.
Not all linguistic innovations need begin with a genetic change in the linguistic abilities of
speakers. Former Secretary of State Alexander Haig achieved notoriety with expressions
such as Let me caveat that or That statement has to be properly nuanced. As listeners we
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 30
cringe at the ungrammaticality but we have no trouble understanding him and would be
hard-pressed to come up with a concise grammatical alternative. The double standard
exemplified by Haigspeak is fairly common in speech (Pinker, 1989b). Most likely this was
always true, and innovations driven by cognitive processes exploiting analogy, metaphor,
iconicity, conscious folk etymology, and so on, if useful enough, could set up pressures for
both speakers and hearers to grammaticize those innovations. Note as well that if a single
mental database is used in production and comprehension (Bresnan and Kaplan, 1982)
evolutionary changes in response to pressure on one performance would automatically
transfer to the other.
5.2.2. Categorical Rules.
Many linguistic rules are categorical, all-or-none operations on symbols (see, e.g.,
Pinker and Prince, 1988, 1989). How could such structures evolve in a gradual
sequence? Bates et al. (1989), presumably echoing Gould's "5% of an eye," (1989)
write:
What protoform can we possibly envision that could have given birth to constraints on the
extraction of noun phrases from an embedded clause? What could it conceivably mean for
an organism to possess half a symbol, or three quarters of a rule? (p. 3) ...monadic symbols,
absolute rules and modular systems must be acquired as a whole, on a yes-or-no basis -- a
process that cries out for a Creationist explanation. (p. 30)"
However, two issues are being collapsed here. While one might justifiably argue that an
entire system of grammar must evolve in a gradual continuous sequence, that does not
mean that every aspect of every rule must evolve in a gradual continuous sequence. As
mentioned, mutant fruit flies can have a full leg growing where an antenna should be and
the evolution of new taxa with different numbers of appendages from their ancestors is
often attributed to such homeotic mutations. No single mutation or recombination could
have led to an entire universal grammar, but it could have led a parent with an n-rule
grammar to have an offspring with an n+1 rule grammar, or a parent with an m-symbol rule
to have an offspring with an m+1 symbol rule. It could also lead to a parent with no
grammatical rules at all and just rote associations to have an offspring with a single rule.
Grammatical rules are symbol-manipulations whose skeletal form is shared by many other
mental systems. Indeed discrete symbol manipulations, free from graded application based
on similarity to memorized cases, is highly useful in many domains of cognition, especially
those involving socially shared information (Freyd, 1983; Pinker and Prince, 1989;
Smolensky, 1988). If a genetic change caused generic copies of a nonlinguistic symbolreplacement operation to pop up within the neural system underlying communication, such
protorules could be put to use as parts of encoding and decoding schemes, whereupon they
could be subject to selective forces tailoring them to specific demands of language. Rozin
(1976) and Shepard (1986) have argued that the evolution of intelligence was made
possible by just such sequences.
5.2.3. Perturbations of Formal Grammars.
Grammars are thought to be complex computational systems with many interacting
rules and conditions. Chomsky (1981) has emphasized how grammars have a rich
deductive structure in which a minor change to a single principle can have dramatic
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 31
effects on the language as a whole as its effects cascade through grammatical
derivations. This raises the question of how the entire system could be viable
under the far more major perturbations that could be expected during evolutionary
history. Does grammar degrade gracefully as we extrapolate backwards in time?
Would a universal grammar with an altered or missing version of some component
be good for anything, or would it result in nothing but blocked derivations, filtered
constructions, and partial structures? Lieberman (1989: 200) claims that "The only
model of human evolution that would be consistent with the current standard
linguistic theory is a sudden saltation that furnished human beings with the neural
bases for language." Similarly, Bates, et al. (1989: 2-3) claim that "If the basic
structural principles of language cannot be learned (bottom up) or derived (top
down), there are only two possible explanations for their existence: either Universal
Grammar was endowed to us directly by the Creator, or else our species has
undergone a mutation of unprecedented magnitude, a cognitive equivalent of the
Big Bang."
But such arguments are based on a confusion. While a grammar for an existing language
cannot tolerate minor perturbations and still be a grammar for a language that a modern
linguist would recognize, that does not mean that it cannot be a grammar at all. To put it
crudely, there is no requirement that the languages of Homo Erectus fall into the class of
possible Homo Sapiens languages. Furthermore language abilities consist of not just formal
grammar but also such nonlinguistic cognitive processes as analogy, rote memory, and
Haigspeak. Chomsky (1981) refers to such processes as constituting the "periphery" of
grammar but a better metaphor may put them in the "interstices," where they would
function as a kind of jerry-rigging that could allow formally incomplete grammars to be
used in generating and comprehending sentences.
The assertion that a natural language grammar either functions as a whole or not at all is
surprisingly common. But it has no more merit than similar claims about eyes, wings, and
webs that frequently pop up in the anti-Darwinian literature (see Dawkins, 1986, for
examples), and which occasionally trigger hasty leaps to claims about exaptation. Pidgins,
contact languages, Basic English, and the language of children, immigrants, tourists,
aphasics, telegrams, and headlines provide ample proof that there is a vast continuum of
viable communicative systems displaying a continuous gradation of efficiency and
expressive power (see Bickerton, 1986). This is exactly what the theory of natural selection
requires.
Our suggestions about interactions between learning and innate structure in evolution are
supported by an interesting simulation of the Baldwin effect by Hinton and Nowlan (1987).
They consider the worst imaginable scenario for evolution by small steps: a neural network
with 20 connections (which can be either excitatory or inhibitory) that conveys no fitness
advantage unless all 20 are correctly set. So not only is it no good to have 5% of the
network; it's no good to have 95%. In a population of organisms whose connections are
determined by random mutations a fitter mutant arises at a rate of only about once every
million (2 ) genetically distinct organisms, and its advantages are immediately lost if the
organism reproduces sexually. But now consider an organism where the connections are
either genetically fixed to one or the other value or are settable by learning, determined by
random mutation with an average of 10 connections fixed. The organism tries out random
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 32
settings for the modifiable connections until it hits upon the combination that is
advantageous; this is recognizable to the organism and causes it to retain those settings.
Having attained that state the organism enjoys a higher rate of reproduction; the sooner it
attains it, the greater the benefit. In such a population there is an advantage to having less
than 100% of the correct network. Among the organisms with, say, 10 innate connections,
the one in every thousand (2 ) that has the right ones will have some probability of attaining
the entire network; in a thousand learning trials, this probability is fairly high. For the
offspring of that organism, there are increasing advantages to having more and more of the
correct connections innately determined, because with more correct connections to begin
with, it takes less time to learn the rest, and the chances of going through life without
having learned them get smaller.
Hinton and Nowlan confirmed these intuitions in a computer simulation, demonstrating
nicely that learning can guide evolution, as the argument in this section requires, by turning
a spike in fitness space into a gradient. Moreover they made an interesting discovery.
Though there is always a selection pressure to make learnable connections innate, this
pressure diminishes sharply as most of the connections come to be innately set, because it
becomes increasingly unlikely that learning will fail for the rest. This is consistent with the
speculation that the multiplicity of human languages is in part a consequence of learning
mechanisms existing prior to (or at least independent of) the mechanisms specifically
dedicated to language. Such learning devices may have been the sections of the ladder that
evolution had no need to kick away.
5.3. Reproductive Advantages of Better Grammars
David Premack (1985: 281-282) writes:
I challenge the reader to reconstruct the scenario that would confer
selective fitness on recursiveness. Language evolved, it is
conjectured, at a time when humans or protohumans were hunting
mastodons. ... Would it be a great advantage for one of our ancestors
squatting alongside the embers, to be able to remark: "Beware of the
short beast whose front hoof Bob cracked when, having forgotten his
own spear back at camp, he got in a glancing blow with the dull spear
he borrowed from Jack"?
Human language is an embarrassment for evolutionary theory because it is
vastly more powerful than one can account for in terms of selective fitness.
A semantic language with simple mapping rules, of a kind one might
suppose that the chimpanzee would have, appears to confer all the
advantages one normally associates with discussions of mastodon hunting or
the like. For discussions of that kind, syntactic classes, structure-dependent
rules, recursion and the rest, are overly powerful devices, absurdly so.
Premack's rhetorical challenge captures a conviction that many people find
compelling, perhaps even self-evident, and it is worth considering why. It is a good
example of what Dawkins (1986) calls the Argument from Personal Incredulity. The
argument draws on people's poor intuitive grasp of probabilistic processes,
especially those that operate over the immensities of time available for evolution.
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 33
The passage also gains intuitive force because of the widespread stereotype of
prehistoric humans as grunting cave men whose main reproductive challenge was
running away from tigers or hunting mastodons. The corollary would seem to be
that only humans in modern industrial societies -- and maybe only academics, it is
sometimes implied -- need to use sophisticated mental machinery. But compelling
as these common-sense intuitions are, they must be resisted.
5.3.1. Effects of small selective advantages.
First one must be reminded of the fact that tiny selective advantages are sufficient
for evolutionary change. According to Haldane's (1924) classic calculations, for
example, a variant that produces on average 1 per cent more offspring than its
alternative allele would increase in frequency from 0.1 per cent to 99.9 per cent of
the population in just over 4,000 generations. Even in long-lived humans this fits
comfortably into the evolutionary timetable. (Needless to say fixations of different
genes can go on in parallel.) Furthermore the phenotypic effects of a beneficial
genetic change need not be observable in any single generation. Stebbins (1982)
constructs a mathematical scenario in which a mouselike animal is subject to
selection pressure for increased size. The pressure is so small that it cannot be
measured by human observers, and the actual increase in size from one
generation to the next is also so small that it cannot be measured against the noise
of individual variation. Nonetheless this mouse would evolve to the size of an
elephant in 12,000 generations, a slice of time that is geologically "instantaneous."
Finally, very small advantages can also play a role in macroevolutionary
successions among competing populations of similar organisms. Zubrow (1987)
calculates that a 1% difference in mortality rates among geographically overlapping
Neanderthal and modern populations could have led to the extinction of the former
within 30 generations, or a single millennium.
5.3.2. Grammatical complexity and technology.
It has often been pointed out that our species is characterized by two features -technology and social relations among nonkin -- that have attained levels of
complexity unprecedented in the animal kingdom. Toolmaking is the most widely
advertised ability, but the knowledge underlying it is only a part of human
technological competence. Modern hunter-gatherers, whose lifestyle is our best
source of evidence for that of our ancestors, have a folk biology encompassing
knowledge of the life cycles, ecology, and behavior of wild plants and animals "that
is detailed and thorough enough to astonish and inform professional botanists and
zoologists" (Konner, 1982: 5). This ability allows the modern !Kung San, for
example, to enjoy a nutritionally complete diet with small amounts of effort in what
appears to us to be a barren desert. Isaacs (1983) interprets fossil remains of
home bases used as evidence for a lifestyle depending heavily on acquired
knowledge of the environment as far back as two million years ago in Homo
Habilis. An oft-noted special feature of humans is that such knowledge can
accumulate across generations. Premack (1985) reviews evidence that pedagogy
is a universal and species-specific human trait, and the usefulness of language in
pedagogy is not something that can be reasonably doubted. As Brandon and
Hornstein (1986) emphasize, presumably there is a large selective advantage
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 34
conferred by being able to learn in a way that is essentially stimulus-free (Williams,
1966, made a similar point.) Children can learn from a parent that a food is
poisonous or a particular animal is dangerous; they do not have to observe or
experience this by themselves.
With regard to adult-to-adult pedagogy, Konner (1982: 171) notes that the !Kung discuss
everything from the location of food sources to the behavior of predators to the movements
of migratory game. Not only stories, but great stores of knowledge are exchanged around
the fire among the !Kung and the dramatizations -- perhaps best of all -- bear knowledge
critical to survival. A way of life that is difficult enough would, without such knowledge,
become simply impossible.
Devices designed for communicating precise information about time, space, predicateargument relations, restrictive modification, and modality are not wasted in such efforts.
Recursion in particular is extraordinarily useful. Premack repeats a common misconception
when he uses tortuous phrases as an exemplification of recursive syntax; without recursion
you can't say the man's hat or I think he left. All you need for recursion is an ability to
embed a phrase containing a noun phrase within another noun phrase or a clause within
another clause, which falls out of pairs of rules as simple as NP --> det N PP and PP --> P
NP. Given such a capacity one can now specify reference to an object to an arbitrarily fine
level of precision. These abilities can make a big difference. For example, it makes a big
difference whether a far-off region is reached by taking the trail that is in front of the large
tree or the trail that the large tree is in front of. It makes a difference whether that region
has animals that you can eat or animals that can eat you. It makes a difference whether it
has fruit that is ripe or fruit that was ripe or fruit that will be ripe. It makes a difference
whether you can get there if you walk for three days or whether you can get there and walk
for three days.
5.3.3. Grammatical complexity and social interactions.
What is less generally appreciated is how important linguistically-supported social
interactions are to a hunter-gatherer way of life. Humans everywhere depend on
cooperative efforts for survival. Isaac (1983) reviews evidence that a lifestyle
depending on social interactions among nonkin was present in Homo Habilis more
than two million years ago. Language in particular would seem to be deeply woven
into such interactions, in a manner that is not qualitatively different from that of our
own "advanced" culture. Konner (1982) writes:
War is unknown. Conflicts within the group are resolved by talking, sometimes half or all
the night, for nights, weeks on end. After two years with the San, I came to think of the
Pleistocene epoch of human history (the three million years during which we evolved) as
one interminable marathon encounter group. When we slept in a grass hut in one of their
villages, there were many nights when its flimsy walls leaked charged exchanges from the
circle around the fire, frank expressions of feeling and contention beginning when the dusk
fires were lit and running on until the dawn. (p. 7)
If what lawyers and judges do is work, then when the !Kung sit up all night at a meeting
discussing a hotly contested divorce, that is also work. If what psychotherapists and
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 35
ministers do is work, then when a !Kung man or woman spends hours in an enervating
trance trying to cure people, that is also work (p. 371).
Reliance on such exchanges puts a premium on the ability to convey socially-relevant
abstract information such as time, possession, beliefs, desires, tendencies, obligations, truth,
probability, hypotheticals, and counterfactuals. Once again, recursion is far from being an
"overly powerful device." The capacity to embed propositions within other propositions, as
in [ He thinks that S] or [ She said that [ he thinks that S]], is essential to the expression of
beliefs about the intentional states of others.
Furthermore, in a group of communicators competing for attention and sympathies there is
a premium on the ability to engage, interest, and persuade listeners. This in turn encourages
the development of discourse and rhetorical skills and the pragmatically-relevant
grammatical devices that support them. Symons' (1979) observation that tribal chiefs are
often both gifted orators and highly polygynous is a splendid prod to any imagination that
cannot conceive of how linguistic skills could make a Darwinian difference.
5.3.4. Social use of language and evolutionary acceleration.
The social value of complex language probably played a profound role in human
evolution that is best appreciated by examining the dynamics of cooperative
interactions among individuals. As mentioned, humans, probably early on, fell into
a lifestyle that depended on extended cooperation for food, safety, nurturance, and
reproductive opportunities. This lifestyle presents extraordinary opportunities for
evolutionary gains and losses. On the one hand it benefits all participants by
surmounting prisoners' dilemmas. On the other hand it is vulnerable to invasion by
cheaters who reap the benefits without paying the costs (Axelrod and Hamilton,
1981; Cosmides, 1989; Hamilton, 1964; Maynard Smith, 1974; Trivers, 1971). The
minimum cognitive apparatus needed to sustain this lifestyle is memory for
individuals and the ability to enforce social contracts of the form "If you take a
benefit then you must pay a cost" (Cosmides, 1989). This alone puts a demand on
the linguistic expression of rather subtle semantic distinctions. It makes a
difference whether you understand me as saying that if you give me some of your
fruit I will share meat that I will get, or that you should give me some fruit because I
shared meat that I got, or that if you don't give me some fruit I will take back the
meat that I got.
But this is only a beginning. Cooperation opens the door to advances in the ability of
cheaters to fool people into believing that they have paid a cost or that they have not taken a
benefit. This in turn puts pressure on the ability to detect subtle signs of such cheating,
which puts pressure on the ability to cheat in less detectable ways, and so on. It has been
noted that this sets the stage for a cognitive "arms race" (e.g., Cosmides and Tooby, 1989;
Dawkins, 1976; Tooby and DeVore, 1987; Trivers, 1971). Elsewhere in evolution such
competitive feedback loops, such as in the struggle between cheetahs and gazelles, have led
to the rapid evolution of spectacular structures and abilities (Dawkins, 1982). The unusually
rapid enlargement of the human brain, especially the frontal lobes, has been attributed to
such an arms race (Alexander, 1987; Rose, 1980). After all, it doesn't take all that much
brain power to master the ins and outs of a rock or to get the better of a berry. But
interacting with an organism of approximately equal mental abilities whose motives are at
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 36
times outright malevolent makes formidable and ever-escalating demands on cognition.
This competition is not reserved for obvious adversaries. Partial conflicts of reproductive
interest between male and female, sibling and sibling, and parent and offspring are inherent
to the human condition (Symons, 1979; Tooby and DeVore, 1987; Trivers, 1974).
It should not take much imagination to appreciate the role of language in a cognitive arms
race. In all cultures human interactions are mediated by attempts at persuasion and
argument. How a choice is framed plays a huge role in determining which alternative
people choose (Tversky and Kahneman, 1981). The ability to frame an offer so that it
appears to present maximal benefit and minimum cost to the buyer, and the ability to see
through such attempts and to formulate persuasive counterproposals, would have been a
skill of inestimable value in primitive negotiations, as it is today. So is the ability to learn of
other people's desires and obligations through gossip, an apparently universal human vice
(Cosmides and Tooby, 1989; Symons, 1979).
In sum, primitive humans lived in a world in which language was woven into the intrigues
of politics, economics, technology, family, sex, and friendship and that played key roles in
individual reproductive success. They could no more live with a Me-Tarzan-you-Jane level
of grammar than we could.
5.4. Phyletic Continuity
Bates et al. (1989), Greenfield (1988), Lieberman (1976, 1984) argue that if
language evolved in humans by natural selection, it must have antecedents in
closely-related species such as chimpanzees, which share 99% of their genetic
material with us and may have diverged from a common ancestor as recently as 57 million years ago (King and Wilson, 1975; Miyamoto, Slightom, and Goodman,
1987). Similarly, since no biological ability can evolve out of nothing, they claim, we
should find evidence of nonlinguistic abilities in humans that are continuous with
grammar. Lieberman claims that motor programs are preadaptations for syntactic
rules while Bates (1976) and Greenfield (Greenfield and Smith, 1976) suggest that
communicative gestures flow into linguistic naming. As Bates et al. (1989: 8) put it,
" ... we have to abandon any strong version of the discontinuity claim that has
characterized generative grammar for thirty years. We have to find some way to
ground symbols and syntax in the mental material that we share with other
species."
The specific empirical claims have been disputed. Seidenberg and Petitto (Seidenberg,
1986; Seidenberg and Petitto, 1979, 1987) have reviewed the evidence of the signing
abilities of apes and concluded that they show no significant resemblance to human
language or to the process of acquiring it. In a study of the acquisition of sign language in
deaf children, Petitto (1987) argues that nonlinguistic gestures and true linguistic names,
even when both share the manual-visual channel, are completely dissociable. These
conclusions could be fodder for the claim that natural language represents a discontinuity
from other primate abilities and so could not have evolved by natural selection.
We find the Seidenberg and Petitto demonstrations convincing, but our argument is not
based on whether they are true. Rather we completely disagree with the premise (not theirs)
that the debate over ape signing should be treated as a referendum on whether human
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 37
language evolved by natural selection. Of course human language, like other complex
adaptations, could not have evolved overnight. But then there is no law of biology that says
that scientists are blessed with the good fortune of being able to find evolutionary
antecedents to any modern structure in some other living species. The first recognizably
distinct mental system that constituted an antecedent to modern human language may have
appeared in a species that diverged from the chimp-human common ancestor, such as
Australopithecus Afarensis or any of the subsequent hominid groups that led to our species.
Moreover chimpanzees themselves are not generalized common ancestors but presumably
have done some evolving of their own since the split. We must be prepared for the possible
bad news that there just aren't any living creatures with homologues of human language,
and let the chimp signing debate come down as it will.
As far as we know this would still leave plenty of time for language to have evolved: 3.5 - 5
million years, if early Australopithecines were the first talkers, or, as an absolute minimum,
several hundred thousand years (Stringer and Andrews, 1988), in the unlikely event that
early Homo Sapiens was the first. (For what it's worth, Broca's area is said to be visible in
cranial endocasts of two million year-old fossil hominids; Falk, 1983; Tobias, 1981.) There
is also no justification in trying to squeeze conclusions out of the genetic data. On the order
of forty million base pairs differ between chimpanzees and humans, and we see no reason
to doubt that universal grammar would fit into these 10 megabytes with lots of room left
over, especially if provisions for the elementary operations of a symbol-manipulation
architecture are specified in the the remaining 99% of the genome (see Seidenberg, 1986,
for discussion).
In fact there is even more scope for design differences than the gross amount of nonshared
genetic material suggests. The 1% difference between chimps and humans represents the
fraction of base pairs that are different. But genes are long stretches of base pairs and if
even one pair is different, the entire functioning product of that gene could be different. Just
as replacing one bit in every byte leads to text that is 100% different, not 12.5% different, it
is possible for the differing base pairs to be apportioned so that 100% of the genes of
humans and chimps are different in function. Though this extreme possibility is, of course,
unlikely, it warns us not to draw any conclusions about phenotypic similarity from degree
of genomic overlap.(Note 4)
As for continuity between language and nonlinguistic neural mechanisms, we find it ironic
that arguments that are touted as being "biological" do not take even the most elementary
steps to distinguish between analogy and homology. Lieberman's claim that syntactic rules
must be retooled motor programs, a putative case of preadaptation, is a good example. It
may be right, but there is no reason to believe it. Lieberman's evidence is only that motor
programs are hierarchically organized and serially ordered, and so is syntax. But
hierarchical organization characterizes many neural systems, perhaps any system, living or
nonliving, that we would want to call complex (Simon, 1969). And an organism that lives
in real time is going to need a variety of perceptual, motor, and central mechanisms that
keep track of serial order. Hierarchy and seriality are so useful that for all we know they
may have evolved many times in neural systems (Bickerton, 1984, 1986, also makes this
point). To distinguish true homology from mere analogy it is necessary to find some unique
derived nonadaptive character shared by the relevant systems, for example, some quirk of
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 38
grammar that can be seen in another system. Not only has no such shared character been
shown, but the dissimilarities between syntax and motor control are rather striking. Motor
control is a game of inches so its control programs must have open continuous parameters
for time and space at every level of organization. Syntax has no such analogue parameters.
A far better case could be made that grammar exploited mechanisms originally used for the
conceptualization of topology and antagonistic forces (Jackendoff, 1983; Pinker, 1989b;
Talmy, 1983, 1988), but that is another story.
6. Conclusion
As we warned, the thrust of this paper has been entirely conventional. All we have
argued is that human language, like other specialized biological systems, evolved
by natural selection. Our conclusion is based on two facts that we would think
would be entirely uncontroversial: language shows signs of complex design for the
communication of propositional structures, and the only explanation for the origin of
organs with complex design is the process of natural selection. Although
distinguished scientists from a wide variety of fields and ideologies have tried to
cast doubt on an orthodox Darwinian account of the evolution of a biological
specialization for grammar, upon close examination none of the arguments is
compelling.
But we hope we have done more than try and set the record straight. Skepticism about the
possibility of saying anything of scientific value about language evolution has a long
history, beginning in the prohibition against discussing the topic by the
Soci@act[h]t@act[h] de Linguistique de Paris in 1866 and culminating in the the
encyclopedic volume edited by Harnad, Steklis, and Lancaster (1976) that pitted a few
daring speculators against an army of doubters. A suspicious attitude is not entirely
unwarranted when one reads about The Age of Modifiers, Pithecanthropus Alalus ("Apeman without speech"), and the Heave-ho theory. But such skepticism should not lead to
equally unsupported assertions about the necessity of spandrels and saltations.
A major problem among even the more responsible attempts to speculate about the origins
of language has been that they ignore the wealth of specific knowledge about the structure
of grammar discovered during the past 30 years. As a result language competence has been
equated with cognitive development, leading to confusions between the evolution of
language and the evolution of thought, or has been expediently equated with activities that
leave tangible remnants, such as tool manufacture, art, and conquest.
We think there is a wealth of respectable new scientific information relevant to the
evolution of language that has never been properly synthesized. The computational theory
of mind, generative grammar, articulatory and acoustic phonetics, developmental
psycholinguistics, and the study of dynamics of diachronic change could profitably be
combined with recent molecular, archeological, and comparative neuroanatomical
discoveries and with strategic modeling of evolution using insights from evolutionary
theory and anthropology (see, e.g., Bickerton, 1981; Brandon and Hornstein, 1986;
Barkow, Cosmides, and Tooby, in press; Hurford, 1989a, 1989b; Tooby & DeVore, 1987;
Hinton and Nowlan, 1987). It is certain there many questions about the evolution of
PINKER & BLOOM
/
NATURAL LANGUAGE AND NATURAL SELECTION 39
language that we will never answer. But we are optimistic that there are insights to be
gained, if only the problems are properly posed.